CN116484095A - Recommended methods, systems, electronic devices, and computer-readable storage media - Google Patents

Abstract

The application discloses a recommendation method, a recommendation system, a recommendation server, an electronic device and a computer readable storage medium, wherein the recommendation method comprises the following steps: constructing a recommendation scheme list, wherein the recommendation scheme list comprises N candidate recommendation schemes, and preset quantity ratios among M categories of preset recommendation schemes in the N candidate recommendation schemes are met; responding to a received recommendation request of a user, and acquiring a first recommendation scheme aiming at the user; querying a first candidate recommended scheme which is the same as the first recommended scheme category in the recommended scheme list; and recommending the first candidate recommended scheme to the user in response to the first candidate recommended scheme being queried in the recommended scheme list, and deleting the first candidate recommended scheme from the recommended scheme list. The method solves the technical problems of large proportion deviation, long balancing time and unexpected problem among various recommended schemes caused by postposition adjustment in the existing recommended method.

Description

Recommended methods, systems, electronic devices, and computer-readable storage media

technical field

The present application relates to the technical field of artificial intelligence, and in particular to a recommendation method, a recommendation server, a recommendation system, electronic equipment, and a computer-readable storage medium.

Background technique

With the development of information technology and the Internet, information overload has become a common challenge faced by both consumers and producers. Based on this, personalized recommendation has emerged. The core of personalized recommendation is to recommend content (such as commodities, information, mechanisms, etc.) that users may be interested in for each user. At present, there are often multiple recommendation schemes in a recommendation activity. In the process of personalized recommendation, controlling the recommendation ratio of each recommendation scheme is a necessary condition for maintaining the long-term economic ecology.

In the existing recommendation methods, the adjustment of the ratio between the recommended schemes is mainly a post-adjustment, that is, after the recommendation activity has been running for a period of time, the floating coefficient is adjusted according to the real proportion of each pushing scheme to balance the real proportion to the preset proportion. After several post-adjustments, although the proportions of the recommended solutions can be balanced after a period of time, the balance time required to reach the balance cannot be predicted, and it is difficult to avoid large deviations in any period of the balance time.

Therefore, the existing recommendation methods have technical problems such as large proportional deviations among recommended schemes, long balancing time and unpredictable technical problems caused by post-adjustment.

Contents of the invention

The present application provides a recommendation method, a recommendation server, a recommendation system, electronic equipment, and a computer-readable storage medium, so as to solve the existing technical problems in existing recommendation methods, such as large ratio deviations among recommended solutions, long balancing time, and unpredictable technical problems caused by post-adjustment.

The embodiment of this application provides a recommended method, which includes:

Constructing a list of recommendation schemes, the list of recommendation schemes includes N candidate recommendation schemes, and among the N candidate recommendation schemes, M preset recommendation schemes satisfy a preset quantity ratio, wherein N and M are both positive integers greater than 1;

Responding to receiving a recommendation request from a user, acquiring a first recommendation plan for the user;

querying the list of recommended solutions for a first candidate recommended solution of the same category as the first recommended solution;

In response to finding the first candidate recommendation solution in the recommendation solution list, recommending the first candidate recommendation solution to the user, and deleting the first candidate recommendation solution from the recommendation solution list.

The embodiment of the present application also provides a recommendation server. The recommendation server includes: a recommended solution list construction unit, a recommended solution acquisition unit, a candidate recommended solution query unit, and a recommendation unit;

The recommendation scheme list construction unit is used to construct a recommendation scheme list, the recommendation scheme list includes N candidate recommendation schemes, and among the N candidate recommendation schemes, the preset recommendation schemes of M categories satisfy a preset quantity ratio, wherein N and M are both positive integers greater than 1;

The recommendation solution obtaining unit is configured to obtain a first recommendation solution for the user in response to receiving a recommendation request from the user;

The candidate recommendation solution query unit is configured to query the first candidate recommendation solution of the same category as the first recommendation solution in the recommendation solution list;

The recommending unit is configured to, in response to finding the first candidate recommendation solution in the recommendation solution list, recommend the first candidate recommendation solution to the user, and delete the first candidate recommendation solution from the recommendation solution list.

The embodiment of the present application also provides a recommendation system, the system includes: a user terminal, a business server, and the above-mentioned recommendation server; wherein,

The client includes: a first receiving module, a first processing module, and a first sending module;

The first receiving module is configured to receive an operation instruction from a user, where the operation instruction is an action that can trigger a recommended behavior;

The first processing module is configured to generate instruction information corresponding to the operation instruction according to the operation instruction;

The first sending module is configured to send the instruction information to the service server;

The business server includes: a second receiving module, a second processing module, and a second sending module;

The second receiving module is configured to receive the instruction information sent by the client;

The second processing module is configured to generate a recommendation request for the user according to the instruction information;

The second sending module is configured to send the recommendation request to the recommendation server;

The recommendation server includes: a third receiving module;

The third receiving module is configured to receive the recommendation request sent by the service server.

The embodiment of the present application also provides an electronic device, including: a memory and a processor;

The memory is used to store one or more computer instructions;

The processor is configured to execute the one or more computer instructions to implement the above method.

The embodiment of the present application also provides a computer-readable storage medium, on which one or more computer instructions are stored, and when the instructions are executed by a processor, the above method is executed.

Compared with the prior art, the recommendation method provided by this application arranges the preset recommendation schemes of M categories according to the preset quantity ratio into a recommendation scheme list including N candidate recommendation schemes, searches the recommendation scheme list for the first candidate recommendation scheme with the same category as the first recommendation scheme for the user to be recommended, recommends the first candidate recommendation scheme to the user to be recommended, and deletes the first candidate recommendation scheme from the recommendation scheme list.

This recommended method has the following advantages:

First, the method recommends to the user to be recommended that the first candidate recommendation scheme is the same candidate recommendation scheme as the first recommendation scheme, and the first recommendation scheme is a recommendation scheme for the user to be recommended. Therefore, this method realizes the user's personalized recommendation.

Second, the first candidate recommendation scheme recommended by the method to the user to be recommended is the candidate recommendation scheme in the recommendation scheme list, and the first candidate recommendation scheme will be deleted from the recommendation scheme list after being recommended, and the recommendation scheme list is formed by arranging the preset recommendation schemes of M categories according to the preset quantity ratio. Therefore, the method realizes strict control of the recommendation ratio of each preset recommendation scheme.

Thirdly, the first candidate recommendation scheme recommended by the method to the user to be recommended is the candidate recommendation scheme in the recommendation scheme list, and the first candidate recommendation scheme will be deleted from the recommendation scheme list after being recommended. Therefore, the method continuously recommends a certain category of preset recommendation schemes.

Fourth, the method recommends to the user to be recommended that the first candidate recommendation solution is a candidate recommendation solution in the recommendation solution list, and the first candidate recommendation solution will be deleted from the recommendation solution list after being recommended. Therefore, by controlling the number N of candidate recommendation solutions included in the recommendation solution list, it is possible to control the time when the recommendation ratio of each category of preset recommendation solutions reaches balance, and realize the predictability of the balance time.

Therefore, the recommendation method provided by the present application solves the technical problems of the existing recommendation method, such as large proportional deviation among recommended solutions, long balancing time and unpredictable technical problems caused by post-adjustment.

Description of drawings

FIG. 1 is an application system diagram of a recommendation method provided in an embodiment of the present application;

Fig. 2 is a flow chart of the recommendation method provided by the first embodiment of the present application;

FIG. 3 is a schematic diagram of a list of recommended solutions provided by the first embodiment of the present application;

Fig. 4 is a schematic diagram of another recommended solution list provided by the first embodiment of the present application;

Fig. 5 is a schematic diagram of another recommended solution list provided by the first embodiment of the present application;

Fig. 6 is a schematic diagram of comparison results of the recommendation methods provided by the first embodiment of the present application;

FIG. 7 is a schematic structural diagram of a recommendation server provided in a second embodiment of the present application;

FIG. 8 is a schematic structural diagram of a recommendation system provided by a third embodiment of the present application;

FIG. 9 is a schematic structural diagram of an electronic device provided by a fourth embodiment of the present application.

Detailed ways

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the application. However, the present application can be implemented in many other ways different from those described here, and those skilled in the art can make similar promotions without violating the connotation of the present application. Therefore, the present application is not limited by the specific implementation disclosed below.

The nouns involved in the embodiments of the present application are introduced below:

Recommendation refers to the process of recommending content (such as products, information, mechanisms, etc.) that the user may be interested in to the user based on the user's characteristic information (such as: the user's personal information, the user's interest characteristics, the user's purchase behavior, the user's historical reading status, etc.) in response to the user's recommendation request. Both recommendation and search are methods to help users quickly find useful information, but recommendation is the process of providing users with recommended content by analyzing their user characteristic information when the user does not provide a clear need; while search is the process of providing users with search content when the user actively provides accurate keywords.

A recommendation system is a personalized information system that recommends products, information, mechanisms, etc. that users are interested in to users. For example: in the game field, the recommendation system can recommend game props suitable for the player to the game player according to the player's characteristic information (such as: game level, prop usage, etc.). Another example: in the field of e-commerce, the recommendation system can recommend products with a high possibility of purchase to users based on user characteristic information (such as: age, region, income level, gender, etc.).

The recommendation algorithm is an algorithm in computer science. Based on the user's characteristic information, through some mathematical algorithms, the content that the user may be interested in is inferred. The recommendation model is a neural network model that implements the recommendation algorithm.

The recommendation model refers to the neural network model constructed according to the user's user characteristic information and user's demand information, which can output personalized recommendation content for the user according to the input user's characteristic information.

With the development of information technology and Internet technology, information overload has become a challenge faced by both consumers and producers. Based on this, personalized recommendation came into being, and the economic benefits increased by personalized recommendation are also increasing. The core of personalized recommendation is to recommend content that the user may be interested in for each user, thereby increasing the user's purchasing power and improving the user's experience. For example: recommend products that users may be interested in for online shopping users, and improve the consumption level of users. Another example: recommend game friends for game players who may be interested in the player, so as to improve the player's retention rate and game experience.

At present, personalized recommendation technology is relatively mature, but it is impossible to have only one recommendation plan in a recommendation activity, and there are often multiple recommendation plans. Controlling the recommendation ratio of each recommendation plan is a necessary condition for maintaining the long-term economic ecology. For example: in game item recommendation activity A, game items with different discount rates are preset. In the short term, game items with lower discount rates will sell better and bring greater economic benefits. However, considering the long-term game economic ecology, it is necessary to strictly control the recommended ratio of the overall discount rate. For example, set the recommended ratio of 10% off game items to 25%, set the recommended ratio of 20% off game items to 50%, and set the recommendation ratio of 30% off game items to 25%. Another example: Items with different rarities and different practicability are preset in the game item recommendation activity B. If more rare and practical items are recommended to players, compared to non-rare and/or narrow-purpose items, the sales of the former will obviously be better. However, considering the long-term game economic ecology, it is necessary to control the distribution ratio of rare props. Therefore, an excellent item recommendation system should not only rely on recommending popular items that are most likely to be purchased by players to improve players’ purchasing power and increase economic benefits, but should match popular items with long-tail items through reasonable gift package matching and user guidance to increase the overall item purchase rate and maintain a better game economic ecology in the long run.

In the existing recommendation method, the adjustment of the ratio between the recommendation schemes is mainly post-adjustment. A floating coefficient will be set for each recommendation scheme in the recommendation activity. After the recommendation activity runs for a period of time, according to the real recommendation proportion of each recommendation scheme, the floating coefficient is adjusted to balance the real recommendation proportion to the preset proportion. If it is low, then the floating coefficient corresponding to recommended scheme A will be increased. In such a way similar to spring oscillation back and forth, after several post-adjustments, although the recommended proportions of each recommended scheme can be balanced (that is, the real recommended proportions of each recommended scheme conform to the preset proportion), a long balance time is required, and the length of the balance time is unpredictable. In addition, although the recommended proportion of each recommended scheme reaches the preset proportion after the balance time, it is difficult to avoid large deviations in any period of the balance time. For the recommendation activities that have strict requirements on the proportion of each recommendation scheme and clear requirements on the balance time, the existing recommendation methods can no longer meet the requirements.

Therefore, how to provide a recommendation method that can reduce the proportional deviation among the recommended solutions and can predict the balance time has become a technical problem to be solved urgently by those skilled in the art.

In view of this, the present application provides a recommendation method, which arranges preset recommendation schemes of M categories into a recommendation scheme list including N candidate recommendation schemes according to a preset quantity ratio, searches the recommendation scheme list for a first candidate recommendation scheme of the same category as the first recommendation scheme for the user to be recommended, recommends the first candidate recommendation scheme to the user to be recommended, and deletes the first candidate recommendation scheme from the recommendation scheme list. This method not only realizes the user's personalized recommendation, but also realizes the strict control of the recommendation proportion among the preset recommendation schemes of each category, reduces the proportion deviation among the preset recommendation schemes of each category, and realizes the predictability of the balance time. The recommendation method provided in this application is applicable to any field that requires a recommendation system, such as the game field, e-commerce field, and the like.

The recommendation method, recommendation server, recommendation system, electronic device, and computer-readable storage medium described in this application will be further described in detail below in conjunction with specific embodiments and accompanying drawings.

FIG. 1 is an application system diagram of a recommendation method provided by an embodiment of the present application. As shown in FIG. 1 , the system includes a client 101 and a server 102 . The client 101 and the server 102 are connected through a network for communication. The user terminal 101 can be a touch terminal, such as a smart phone, a tablet computer, a personal digital assistant (Personal Digital Assistant, PDA) and other devices; it can also be a computer terminal, such as a notebook computer, a desktop computer and the like, and there can be one or more. The server 102 is used to deploy the recommended method provided in this application. The user initiates a recommendation request through the client 101 (such as logging into a shopping website, logging into a game account, etc.), and the recommendation request is transmitted to the server 102 through the network, and the server 102 generates a recommendation for the user through the recommendation method provided in this application, and transmits the recommendation to the client 101 through the network. The server 102 may be a recommendation device for the client 101, or may be a server deployed with the recommendation method provided by this application, and the server may be an independent server, or may be a server cluster composed of multiple sub-servers, wherein each sub-server deploys a module of the recommendation method provided by this application. For example: a recommendation plan list building module, a recommendation plan generation module, etc., can also be a cloud server, and the recommendation method provided by this application is deployed on the cloud server, and provides recommendation services for multiple client terminals 101 at the same time.

The first embodiment of the present application provides a recommendation method, which can be applied to a recommendation server of a recommendation system.

Fig. 2 is a flow chart of the recommendation method provided by this embodiment. The recommendation method provided in this embodiment will be described in detail below with reference to FIG. 2 . The embodiments involved in the following description are used to explain the technical solutions of the present application, and are not intended to limit actual use.

As shown in FIG. 2, the recommendation method provided in this embodiment includes the following steps S201 to S204.

Step S201, constructing a list of recommended solutions, which includes N candidate recommended solutions, and among the N candidate recommended solutions, M categories of preset recommended solutions satisfy a preset quantity ratio, where N and M are both positive integers greater than 1.

The preset recommendation scheme is a recommendation scheme set by the operator for a recommendation activity. For example, the e-commerce operator sets three coupons for a certain discount coupon recommendation activity, which are 10% off coupons, 20% off coupons, and 30% off coupons. These three coupons are the default recommendation schemes for the recommendation activity. Another example: the game operator has set up three types of gift packs for a game gift pack recommendation event, which are ordinary equipment enhancement material gift packs, rare equipment enhancement material gift packs, and epic equipment enhancement material gift packs. These three game gift packs are the default recommendation schemes for this recommendation event.

The preset recommendation plan is determined by the operator during the design stage of the recommendation activity. The operator uses certain historical data as the basis and is guided by the activity design goal. The content to be recommended for the determined recommendation activity does not depend on the precise algorithm. For example, if an e-commerce operator designs a coupon recommendation activity for Double Eleven, then the e-commerce operator can use the real consumption situation of users during the Double Eleven period of the previous year, the actual usage of coupons by users, and the real consumption level of users in the near future as the basis for the historical data. Another example: a game operator designs an equipment enhancement material gift pack recommendation activity for a certain game. Then, the game operator can use recent (such as: 1 month, or 3 months, or 6 months before the recommended event, etc.) players’ real purchases of equipment enhancement materials and players’ actual equipment enhancements as historical data. , Rare Equipment Enhancement Material Gift Pack, and Epic Equipment Enhancement Material Gift Pack. Among them, the normal equipment enhancement material gift pack contains 1 long-tail material, the rare equipment enhancement material gift pack contains 2 long-tail materials, and the epic equipment enhancement material gift pack contains 3 long-tail materials. If the player expects to purchase epic equipment enhancement material gift packs to quickly strengthen props, they will purchase 3 long-tail materials at the same time, increasing the purchase rate of long-tail materials.

In this embodiment, the number of categories of preset recommendation schemes set by the operator for the recommendation activity is defined as M, and the number of preset recommendation schemes in a recommendation activity is at least 1, but the recommendation method provided in this embodiment is the recommendation for M preset recommendation schemes with a preset quantity ratio, therefore, M is a positive integer greater than 1. In actual recommendation activities, the category number of the preset recommendation scheme is also greater than 1. For example: the preset recommendation scheme included in the game item recommendation activity designed for a certain game can cover all the items set in the game, and a single item can be used as a recommendation program, or a combination of multiple items can be used as a recommendation program. Another example: the preset recommendation scheme included in the video recommendation activity designed for a certain video push platform can cover all the videos to be pushed on the video push platform. Since the amount of videos to be pushed is too large, the same type of video can be used as a recommendation scheme. There may be multiple videos to be pushed in these recommendation schemes, and there may be overlap between the videos to be pushed, and the specific implementation method is not limited here.

The candidate recommendation scheme is a recommendation scheme that can be recommended to the user, and is a constituent unit of the recommendation scheme list. In this embodiment, arranging the preset recommendation schemes in the recommendation scheme list according to the preset quantity ratio becomes the candidate recommendation scheme. For example, the preset recommendation schemes include scheme X, scheme Y, and scheme Z. In this embodiment, only by arranging the preset recommendation schemes into a recommendation scheme list can the candidate recommendation schemes in the recommendation scheme list be recommended to the user.

The preset quantity ratio is the occurrence ratio of each category of preset recommendation schemes designed by the operator in the recommendation activity design stage, that is, the proportion of each preset recommendation scheme that is expected to be actually recommended to users. The preset quantity ratio can be determined by the operator based on the grasp of macro data, combined with the design goals of the recommended activities, expectations for the future, etc., and is not calculated by precise algorithms.

The long-term economic ecology can only be maintained by recommending the preset recommendation schemes of each category to users according to the preset quantity ratio. Take a game operator designing an equipment enhancement material gift pack recommendation activity for a certain game as an example. The game operator has set up three types of preset recommendation schemes in this equipment enhancement material gift pack recommendation activity. Scheme X is a general equipment enhancement material gift pack, scheme Y is a rare equipment enhancement material gift pack, and scheme Z is an epic equipment enhancement material gift pack. The virtual value (equipment upgrade speed) of plan X, plan Y, and plan Z increases sequentially, and the selling price also increases sequentially. In the game, it usually happens that although the price of plan Z is more expensive, because the output in the game is controlled (for example, the dungeon that produces epic equipment enhancement materials has a long cooling time, and the epic equipment materials sold in the store have a purchase limit, and they also need to wait for cooling), for high VIP players, there will be situations where they can buy as much epic equipment enhancement materials as they can buy. At this time, if a large number of recommended plan Z, although it can quickly increase game revenue in a short period of time, it will relatively increase the player's equipment too quickly. In the game environment, the game operator designs expected values for players’ equipment levels (for example, how long does it take for players in a certain echelon to upgrade common equipment to top-level equipment), and the expected values of equipment levels are related to plans such as game version updates. In the above cases, it is necessary to control the recommended ratio of plan Z, that is, to control the number of epic equipment enhancement materials that players have, so as to limit the speed of player equipment upgrades. Therefore, the preset quantity ratio is set by the game operator considering the impact of each category of preset recommendation schemes on the game economic system and the improvement of the player's numerical ability. For example, the preset quantity ratio of scheme X is set to 25%, the preset quantity proportion of scheme Y is set to 50%, and the preset quantity proportion of scheme Z is set to 25%.

The recommended plan list is a plan lineup composed of N candidate recommended plans, and also a plan lineup formed by multiplying integer multiples of M categories of preset recommended plans according to a preset number ratio. That is to say, each candidate recommendation solution in the recommendation solution list belongs to one of the M categories of preset recommendation solutions. For example, if the preset recommendation scheme includes scheme X, scheme Y, and scheme Z, then any candidate recommendation scheme in the recommended scheme list is either scheme X, scheme Y, or scheme Z.

In this embodiment, the number of candidate recommendation solutions included in the recommendation solution list is defined as N, and N candidate recommendation solutions are formed by amplifying preset recommendation solutions of M categories according to a preset quantity ratio. The number of candidate recommendation solutions included in the recommendation solution list can also be understood as the length of the recommendation solution list, and the length of the recommendation solution list is usually determined by the operator according to the expected balance time between preset recommendation solutions of each category, and the historical or expected acquisition times of recommendation requests within the balance time. For example, for a certain game, all players will send 10,000 recommendation requests per hour, and the game operator expects to ensure that there is no significant deviation in the proportion of preset recommendation solutions in each category per hour. Then, the length of the recommended solution list can be set to about 10,000. Another example: For a certain product, users trigger recommendation requests 400 times per minute on average, and the operator expects the balance time of preset recommendation solutions for each category to be measured in minutes. Then, the length of the recommended solution list constructed can be set to about 400.

This embodiment provides an optional method for building a list of recommended solutions, including the following steps S201-1 to S201-4.

Step S201-1, acquiring preset recommendation schemes of the M categories.

In the design stage of recommendation activities, the operator determines the content to be recommended in the recommendation activities based on certain historical data and guided by the design goals of the activities, and forms a number of preset recommendation schemes of different categories. The operator can import all categories of preset recommendation schemes into the recommendation server, so that the recommendation server can obtain all categories of preset recommendation schemes. In this embodiment, the number of categories of preset recommendation schemes is defined as M.

For example: in a game item recommendation activity, the game operator has set up three types of preset recommendation plans, namely plan X, plan Y, and plan Z. Among them, plan X is a combination gift package of item 1 and item 2, plan Y is a combination gift package of item 3 and item 4, and plan Z is a combination gift package of item 5 and item 6. Set the preset quantity ratio of scheme X to 25%, the preset quantity proportion of scheme Y to 50%, and the preset quantity proportion of scheme Z to 25%. The game operator imports the content and preset quantity ratio of plan X, plan Y, and plan Z into the recommendation server.

Step S201-2: Determine the number N of candidate recommendation solutions according to the preset balance time for the preset recommendation solutions of the M categories and the number of acquisitions of the recommendation requests within the preset balance time.

The preset balancing time is a unit time period that represents the consumption of the candidate recommendation solutions in the recommendation solution list. After all the candidate recommendation schemes in the recommendation scheme list are consumed, there must be no significant deviation between the real recommendation ratios of the M categories of preset recommendation schemes included in the recommendation scheme list and the preset quantity ratio, that is, a balance has been reached between the M categories of preset recommendation schemes. Therefore, in this embodiment, the duration of consuming all the candidate recommendation schemes in the recommendation scheme list is defined as the preset balance time. The preset balance time is a parameter set in advance by the operator according to expectations. For example, if the operator expects to ensure that there is no significant deviation between the actual recommendation ratio and the preset quantity ratio of each category of preset recommendation schemes in units of hours, then the preset balance time can be set to 1 hour. Another example: the operator expects to use minutes as the unit of measurement to ensure that the preset recommendation schemes of each category are balanced, then the preset balance time can be set to 1 minute.

The number of acquisitions of the recommendation request within the preset balance time may be determined according to the activity level of the user within the historical period. For example: an e-commerce operator designs a coupon recommendation activity for Double Eleven. The e-commerce operator determines the number of recommendation request acquisitions set in this year’s Double Eleven coupon recommendation activity according to the activity level of users during the Double Eleven period of the previous year, that is, the total number of times that all users trigger recommendation requests (logging in to online shopping pages). Assuming that the average number of recommendation requests triggered by all users within 1 hour during the Double Eleven period of the previous year is 20,000 times, and the preset balance time set in this year’s Double Eleven coupon recommendation activity is 1 hour, then the number of times the recommendation requests are obtained within the preset balance time It can be 20,000 times. Of course, user data in the historical period of this year (such as: October) can also be integrated to predict more accurately the number of times the recommendation request is obtained within the preset balance time.

The operator imports the preset balance time and the number of acquisitions of the recommendation request within the preset balance time as parameters into the recommendation server. The recommendation server can determine the number of candidate recommendation solutions according to the preset balance time and the number of acquisitions of the recommendation request within the preset balance time. In this embodiment, the number of candidate recommendation solutions in the recommendation solution list is defined as N.

In an optional implementation manner of this embodiment, N is one or more times the acquisition times. Continue to illustrate with the above example, assuming that the game operator sets the preset balance time to 1 minute, and the number of acquisitions of the recommendation request within the determined preset balance time is 40, then the game server can determine that the number N of candidate recommendation solutions in the recommendation solution list for the game item recommendation activity is 40, that is, the length of the recommendation solution list is 40. Of course, the number N of candidate recommendation solutions in the recommendation solution list can also be determined as 80, 120, or 160 according to parameters such as the expected expansion information of this recommendation activity input by the game operator.

Step S201-3, determining the amplification factor according to the number N of the candidate recommendation proposals and the preset quantity ratio.

After determining the number of candidate recommendation solutions in the recommendation solution list, the amplification factor can be determined according to the preset quantity ratio, that is, how many times the preset recommendation solutions of each category are increased to form the recommendation solution list.

Using the above example to continue to illustrate, the number N of candidate recommendation schemes in the recommendation scheme list is 40, the preset quantity ratio of scheme X is 25%, the preset quantity proportion of scheme Y is 50%, and the preset quantity proportion of scheme Z is 25%. Then, the ratio of the least common divisor corresponding to scheme X, scheme Y, and scheme Z is [1:2:1]. That is to say, the minimum recommended scheme group includes 1 scheme X, 2 schemes Y, and 1 scheme Z. When there are 40 candidate recommendations, the amplification factor is 10.

Certainly, if the number of candidate recommendation schemes in the determined recommendation scheme list is not an integral multiple of the number of candidate recommendation schemes included in the minimum recommendation scheme group, then the number of candidate recommendation schemes in the recommendation scheme list can be appropriately increased or decreased, such as: the number N of candidate recommendation schemes in the determined recommendation scheme list is 40, but the minimum recommendation scheme group includes 6 candidate recommendation schemes (the proportion of the least common divisor corresponding to scheme X, scheme Y, and scheme Z is [2:3:1]), then, the candidate recommendation schemes in the recommendation scheme list can be The number N of is adjusted from 40 to 42 or 36, so that the preset recommendation schemes of each category can exist in the recommendation scheme list with a preset quantity ratio.

Step S201-4, amplifying the preset recommendation schemes of each category according to the amplification factor to form the recommendation scheme list.

After the expansion multiple is determined, the preset recommendation schemes of each category can be expanded to form a recommendation scheme list, and all the recommendation schemes included in the recommendation scheme list are candidate recommendation schemes.

Continue to explain with the above example, after confirming that the amplification factor is 10, the scheme X, scheme Y, and scheme Z are all amplified according to 10 times, the scheme X is amplified to 10, the scheme Y is amplified to 20, and the scheme Z is amplified to 10. Arrange 10 plans X, 20 plans Y, and 10 plans Z to form a list of recommended plans including 10 plans X, 20 plans Y, and 10 plans Z, which can be expressed as [X, ..., Y, ..., Z, ...]. Each plan X, plan Y, and plan Z in the list of recommended plans is a candidate recommendation plan and can be recommended to players.

FIG. 3 is a schematic diagram of a list of recommended solutions provided in this embodiment.

As shown in Figure 3, in a game item recommendation activity, the game operator’s preset recommendation schemes include scheme X, scheme Y, and scheme Z. Among them, the preset quantity ratio of scheme X is 25%, the preset quantity proportion of scheme Y is 50%, and the preset quantity proportion of scheme Z is 25%. The proportions of scheme X, scheme Y, and scheme Z in the recommended scheme list conform to the preset quantity ratio. Wherein, the arrangement sequence of the X scheme, the Y scheme and the Z scheme is not limited.

It should be noted that, for a recommendation activity corresponding to a business application, the recommendation server will construct a list of recommendation solutions. Whether it is the same recommendation activity corresponding to different business applications or different recommendation activities corresponding to the same business application, the recommendation server will construct the corresponding recommendation solution list. Therefore, there is a one-to-one correspondence between the recommendation solution list, business applications and recommendation activities.

Step S202, in response to receiving a recommendation request from a user, acquiring a first recommendation scheme for the user.

The recommendation request may be request information generated by the user performing a request operation, such as: the user enters "XX product" on a search page of a certain website through the user terminal and clicks the search button; it may also be request information generated by the user performing other operations, such as: the user logs in to a certain game account through the user terminal.

The recommendation request can be directly sent from the client to the recommendation server, or can be forwarded to the recommendation server through an intermediate node. For example, the user performs an operation of logging in to a game account through the client. After the processing module of the client detects this operation, a recommendation request is generated and sent to the recommendation server. Another example: the user performs an operation of logging in to a game account through the client terminal. After the processing module of the client terminal detects the operation, it generates user action information and sends the action information to the game server. The game server responds to the action information, generates a recommendation request, and sends it to the recommendation server. The specific implementation form depends on the specific situation of the business application, and there is no limitation here.

In an optional implementation provided in this embodiment, the recommendation request includes user characteristic information of the user, and the method for obtaining a recommendation plan includes the following steps S202-1 to S202-3.

Step S202-1, inputting the recommendation request of the user into a preset recommendation model, so that the recommendation model determines a plurality of recommendation solutions for the user according to the user feature information in the recommendation request.

The user feature information includes user status information and user behavior information, wherein the user status information is personal information irrelevant to specific business applications, such as: user age, region, income level, gender, etc. The user's status information has a certain guiding significance in the recommendation algorithm. For example, the probability of recommending women's clothing to male users is relatively small, and the probability of recommending ultra-thick cotton clothing to southern users is relatively small. The user's behavior information refers to the user's participation in specific business applications, such as: the user's purchase behavior, the user's historical browsing status, etc. This information is related to specific business applications. For example, for games, user behavior information may include user skill usage, user scores, user item purchases, etc.; for video applications, user behavior information may include user video viewing, user video likes, etc. User behavior information also has certain guiding significance in the recommendation algorithm. For example, the probability of recommending daggers to users who use laser guns more frequently will be lower, and the probability of recommending lifestyle videos to users who like funny videos more times will be lower.

The recommendation model can be a module deployed on the recommendation server, or an independent server. After the recommendation server inputs the recommendation request including user characteristic information into the recommendation model, the recommendation model can output a personalized recommendation plan for the user according to the user characteristic information of the user. Usually there are multiple recommendation schemes, and the recommendation model will also assign a predicted user acceptance probability to each recommendation scheme. For commodities, the user acceptance probability is the user’s purchase probability. For example, the user acceptance probability of recommendation scheme 1 is 0.85, the user acceptance probability of recommendation scheme 2 is 0.12, and the user acceptance probability of recommendation scheme 3 is 0.40. The recommendation model often outputs one or more high-probability recommendation schemes in sequence according to the user acceptance probability of each recommendation scheme.

The method of using the recommendation model to output the recommendation plan is relatively mature. The recommendation model is a neural network model that outputs the user's personalized recommendation plan according to the user's characteristic information. The specific architecture and processing methods of the recommendation model will not be described in detail here.

Step S202-2, sorting the multiple recommendation schemes according to the expected virtual value from high to low.

The expected virtual value is a reference value for measuring the income that each recommendation scheme can bring, and it can increase the recommendation income more than simply measuring the recommendation scheme by the predicted user acceptance probability. For commodities, an optional calculation method of expected virtual value is:

Expected virtual value = purchase probability × commodity virtual price

The recommendation server can sort multiple recommendation schemes according to their expected virtual values. For example, an item recommendation model outputs five recommendation schemes for player A based on player characteristic information, namely: item 1, item 2, item 3, item 4, and item 5. The predicted purchase probabilities of the five recommendation programs are: 0.50, 0.1, 0.25, 0.05, and 0.1. The prices of the five items in the game are: 10 yuan, 5 yuan, 3 yuan, 4 yuan, and 1 yuan. Then, the expected virtual value of each recommended scheme is shown in Table 1:

Table 1 Sorting list of recommended schemes

sequence number Recommended plan purchase probability Virtual price (yuan) expected fictitious value 1 Props 1 0.50 10 5 2 prop 3 0.25 3 0.75 3 Props 2 0.1 5 0.5 4 prop 4 0.05 4 0.2 5 prop 5 0.1 1 0.1

It can be seen from Table 1 that the expected virtual value of item 1 is far greater than that of other items. Although the predicted purchase probability of item 5 is greater than that of item 4, the price of item 4 is higher, which makes the expected virtual value of item 4 exceed the expected virtual value of item 5.

Step S202-3, outputting the first recommendation solution from the plurality of recommendation solutions sorted from high to low of the expected virtual value.

After the recommendation server sorts each recommendation plan according to the expected virtual value, it can output any recommendation plan according to the ranking order. In this embodiment, the output recommendation plan is defined as the first recommendation plan.

In an optional implementation manner provided in this embodiment, the first recommendation scheme is the recommendation scheme with the highest expected virtual value among the multiple recommendation schemes. The outputting the first recommendation solution from the plurality of recommendation solutions sorted from high to low by the expected virtual value includes: selecting the recommendation solution with the highest expected virtual value from the plurality of recommendation solutions as the first recommendation solution output.

Although the recommendation model can calculate multiple recommendation schemes based on the input user characteristic information, there is often only one final recommendation scheme output, that is, the recommendation scheme with the highest user acceptance probability, or user purchase probability, or expected virtual value. In this embodiment, the recommendation server receives the recommendation request from the user, and the first recommendation plan for the user acquired for the first time is the recommendation plan with the highest expected virtual value, such as item 1 in Table 1.

Step S203, querying the list of recommended proposals for a first candidate recommended proposal of the same category as the first recommended proposal.

In the prior art, after the recommendation server obtains the first recommendation plan for the user, it will directly recommend the first recommendation plan to the user. However, in the recommendation method provided in this embodiment, the recommendation server will check whether there is a candidate recommendation plan of the same category as the first recommendation plan in the recommendation list according to the first recommendation plan. In this embodiment, any one of the candidate recommendation schemes in the recommendation scheme list that is of the same category as the first recommendation scheme is defined as the first candidate recommendation scheme.

As shown in Figure 3, the list of recommended schemes includes 10 schemes X, 20 schemes Y, and 10 schemes Z. When the recommendation server receives the user’s recommendation request and acquires the first recommendation plan A for the user, it will check whether there is a candidate recommendation plan of the same category as the first recommendation plan A in the recommendation list.

Step S204, in response to finding the first candidate recommendation solution in the recommendation solution list, recommending the first candidate recommendation solution to the user, and deleting the first candidate recommendation solution from the recommendation solution list.

When the recommendation server finds the first candidate recommendation plan in the list of recommendation plans that has the same category as the first recommendation plan, it recommends the first candidate recommendation plan to the user and deletes the first candidate recommendation plan from the list of recommendation plans.

Fig. 4 is a schematic diagram of another recommended solution list provided by this embodiment.

Take the recommended scheme list shown in FIG. 3 as an example, which includes 10 schemes X, 20 schemes Y, and 10 schemes Z. When the recommendation server receives the user’s recommendation request and obtains the first recommendation plan A for the user, it will check whether there is a candidate recommendation plan in the list of recommendation plans that is the same category as the first recommendation plan A. If plan X is of the same category as the first recommendation plan A, and plan X is included in the list of recommendation plans at this time, then the recommendation server will take any one of plan X as the first candidate recommendation plan, take it out of the list of recommendation plans, and return it to the user. At this time, the list of recommendation plans will be as shown in Figure 4, including nine proposals. X, 20 plans Y, and 10 plans Z.

In this way, continuously receiving recommendation requests from the same user or different users, the recommendation server will continuously take out candidate recommendation solutions from the list of recommendation solutions and return them to the user. Since the process of extracting candidate recommendation solutions is also a process of deleting candidate recommendation solutions from the recommendation solution list, the number of candidate recommendation solutions in the recommendation solution list will gradually decrease with the operation of the recommendation activity. Usually, there will be two situations. One is that one or more types of candidate recommendation solutions in the recommendation solution list are all taken out. X, 20 plans Y, and 10 plans Z have all been taken out and returned to the user, and plan X, plan Y, and plan Z in the recommended plan list are all consumed. The recommended methods for both cases are described below.

First, if the recommendation server does not find a first candidate recommendation solution of the same category as the first recommendation solution in the recommendation solution list, the recommendation method provided in this embodiment further includes the following steps S11-S13.

Step S11 , in response to not finding the first candidate recommendation solution in the list of recommendation solutions, acquiring a second recommendation solution for the user, wherein the second recommendation solution is of a different category from the first recommendation solution.

The second recommendation scheme is any one of the multiple recommendation schemes output by the recommendation model except the first recommendation scheme. In an optional implementation manner, the second recommendation scheme is any recommendation scheme whose expected virtual value is less than the first recommendation scheme. In another optional implementation manner, the second recommendation scheme is the first recommendation scheme whose expected virtual value is arranged after the first recommendation scheme, that is, the expected virtual value of the second recommendation scheme is only smaller than the expected virtual value of the first recommendation scheme. As shown in Table 1, if Item 1 is the first recommended solution, then Item 3 is the second recommended solution.

If the recommendation server does not find the first candidate recommendation plan of the same category as the first recommendation plan in the recommendation plan list, it means that all candidate recommendation plans in the recommendation plan list that are of the same category as the first recommendation plan have been taken out, then the recommendation server will re-acquire the second recommendation plan for the user, for example: plan B, the expected virtual value of plan B is less than the expected virtual value of plan A.

Step S12, querying the list of recommended solutions for second candidate recommended solutions of the same category as the second recommended solution.

After the recommendation server obtains the second recommendation plan, it repeats the query operation, and searches the list of recommendation plans for second candidate recommendation plans of the same category as the second recommendation plan. In this embodiment, any candidate recommendation plan in the list of recommendation plans that is the same as the category of the second recommendation solution but different from the category of the first candidate recommendation solution is defined as the second candidate recommendation solution.

Of course, if there is still no second candidate recommendation solution of the same category as the second recommendation solution in the list of recommendation solutions, the recommendation server will repeat step S11 to obtain a third recommendation solution for the user, and the category of the third recommendation solution is different from that of the first recommendation solution and the second recommendation solution. In an optional implementation manner, the third recommendation scheme is any one of the recommended schemes whose expected virtual value is smaller than the first recommended scheme and the second recommended scheme. In another optional implementation manner, the second recommended scheme is the first recommended scheme whose expected virtual value is arranged after the first recommended scheme, and the third recommended scheme is the second recommended scheme whose expected virtual value is arranged after the first recommended scheme, that is, the expected virtual value of the third recommended scheme is only smaller than the expected virtual value of the second recommended scheme. After the recommendation server acquires the third recommendation solution, it will repeat step S12 to search the list of recommendation solutions for candidate recommendation solutions of the same category as the third recommendation solution. This is repeated until no candidate recommendation scheme can be found in the recommendation scheme list.

Step S13, in response to finding the second candidate recommendation solution in the recommendation solution list, recommending the second candidate recommendation solution to the user, and deleting the second candidate recommendation solution from the recommendation solution list.

The recommendation server finds a second candidate recommendation plan of the same category as the second recommendation plan in the list of recommendation plans, recommends the second candidate recommendation plan to the user, and deletes the second candidate recommendation plan from the list of recommendation plans.

Fig. 5 is a schematic diagram of another recommended solution list provided by this embodiment.

Take the recommended scheme list shown in FIG. 3 as an example, which includes 10 schemes X, 20 schemes Y, and 10 schemes Z. With the operation of the recommendation activity, the recommendation server has taken out 10 plan X, 11 plan Y, and 3 plan Z in the list of recommended plans and returned them to multiple users, that is, plan X no longer exists in the list of recommended plans, and the list of recommended plans includes 9 plans Y and 7 plans Z. When the recommendation server receives the user’s recommendation request and obtains the first recommendation plan A for the user, it will check whether there is a first candidate recommendation plan in the list of recommendation plans. The recommendation server further obtains the second recommendation plan B for the user, and then inquires whether there is a second candidate recommendation plan in the list of recommendation plans that has the same category as the second recommendation plan B. If the category of plan Z is the same as that of the second recommendation plan B, and the list of recommendation plans includes plan Z at this time, then the recommendation server will take out a plan Z from the list of recommended plans and return the plan Z to the user. At this time, the list of recommended plans will be as shown in Figure 5.

Second, if the recommendation server does not find a first candidate recommendation plan of the same category as the first recommendation plan in the recommendation plan list, and does not find any candidate recommendation plan in the recommendation plan list, the recommendation method provided by this embodiment also includes the following steps S21-S23.

Step S21, in response to not finding the first candidate recommendation solution in the list of recommendation solutions, and not finding any of the candidate recommendation solutions in the list of recommendation solutions, generating a second recommendation solution list, wherein the preset recommendation solutions of M categories in the second recommendation solution list satisfy a preset quantity ratio.

If the recommendation server does not find any candidate recommendation solution in the recommendation solution list, it means that all the candidate recommendation solutions in the recommendation solution list have been consumed, then the recommendation server will regenerate the second recommendation solution list. It should be noted that, in the second recommendation list, the preset recommendation schemes of the M categories still need to meet the preset quantity ratio. The length of the second recommendation plan list can be the same as the length of the recommendation plan list, and can also be different from the length of the recommendation plan list. For example: the recommendation plan list is based on the preset quantity ratio of M categories of preset recommendation schemes. The preset recommendation schemes of M categories are expanded by 40 times to form a recommendation scheme queue, and the second recommendation scheme list is based on the preset quantity ratio of M categories of preset recommendation schemes. The preset recommendation schemes of M categories are expanded by 50 times to form a recommendation scheme queue.

In an optional implementation manner provided by this embodiment, the length of the second recommended solution list is the same as that of the recommended solution list, that is, the second recommended solution list is the same as the recommended solution list. Usually, for the same recommendation activity of the same service application, no matter how many recommendation solution lists are regenerated, each recommendation solution list will be the same as the original recommendation solution list.

The original recommendation plan list is a list of recommendation plans constructed by the recommendation server based on the imported preset recommendation plan, preset quantity ratio, preset balance time, and the number of acquisitions of recommendation requests within the preset balance time during the design stage of the recommendation activity.

For example: if the recommended solution list shown in FIG. 3 is the original recommended solution list, then the second recommended solution list also consists of 10 solution X, 20 solution Y, and 10 solution Z.

Step S22, querying the second recommendation plan list for a third candidate recommendation plan of the same category as the first recommendation plan.

The recommendation server continues to search for candidate recommendation solutions of the same category as the first recommendation solution in the second recommendation solution list. In this embodiment, any one of the candidate recommendation solutions in the second recommendation solution list that is of the same category as the first recommendation solution is defined as the third candidate recommendation solution.

Step S23, in response to finding the third candidate recommendation solution in the second recommendation solution list, recommending the third candidate recommendation solution to the user, and deleting the third candidate recommendation solution from the second recommendation solution list.

The recommendation server finds a third candidate recommendation plan of the same category as the first recommendation plan in the second recommendation plan list, recommends the third candidate recommendation plan to the user, and deletes the third candidate recommendation plan from the second recommendation plan list.

If the recommendation scheme list shown in FIG. 3 is the original recommendation scheme list, then after one recommendation, the second recommendation scheme list will be as shown in FIG. 4 .

With the running of the recommendation activity, all candidate recommendation solutions in the second recommendation solution list will be consumed, and the recommendation server will regenerate the third recommendation solution list, and repeat the above steps until the recommendation activity ends.

To sum up, compared with the existing post-adjustment recommendation method, the recommendation method provided by this embodiment belongs to pre-adjustment. Firstly, a recommendation list is constructed according to the preset quantity ratio, preset balance time, and number of acquisition requests within the preset balance time of various types of preset recommendation solutions, and then candidate recommendation solutions are selected from the list of recommendation solutions to be recommended to users.

FIG. 6 is a schematic diagram of comparison results between the existing recommendation method provided in this embodiment and the recommendation method provided in this embodiment.

As shown in Figure 6, the abscissa (X) is the running time of a certain recommendation activity, and the ordinate (Y) is the deviation variance between the real recommendation ratio and the preset quantity ratio of each preset recommendation scheme at each running time. Because the existing recommendation method is a post-adjustment, the recommendation model outputs the recommendation plan according to the inherent parameters, monitors the real recommendation ratio of each recommendation plan through the hysteresis, performs reverse adjustment, and balances the ratio back through the subsequent recommendation plan. In such a way similar to the oscillation of a spring back and forth, after several post-adjustments, although the recommended proportions of each preset recommendation scheme can also be balanced (that is, the actual recommended proportion of each preset recommendation scheme conforms to the preset quantity ratio), it requires a long balance time, and the length of the balance time is unpredictable. Because the recommendation method provided in this embodiment is a pre-adjustment, a list of recommended solutions that strictly conforms to the preset quantity ratio is established first, and then candidate recommended solutions are taken out from the list of recommended solutions. Therefore, a jagged deviation curve 602 is presented. It can be seen from the deviation curve 601 and the deviation curve 602 that the recommendation method provided by this embodiment can reduce the proportional deviation among the preset recommendation schemes, and can control the balance time among the preset recommendation schemes.

In general, the longer the recommended plan list is, the greater the maximum deviation will be, but it will not exceed the total number of candidate recommended plans with the largest proportion in the recommended plan list. For example, if the recommended plan list includes 10 plan X, 20 plan Y, and 10 plan Z, then the maximum deviation that can occur is at most 20 (that is, plan Y is recommended to 20 users in a row), and then it will be balanced back immediately.

The shorter the length of the recommendation plan list, the smaller the maximum deviation and the shorter the balance time will be. For example, for a certain game, all players will send 10,000 recommendation requests per hour. If the length of the recommendation plan list is set to 10,000, then the balance can be reached in 1 hour at most. If the length of the recommendation plan list is set to 20,000, it may take 2 hours to reach the balance. That is to say, all the candidate recommendation schemes in the recommendation scheme list are recommended, and the recommendation ratio among the preset recommendation schemes must be balanced.

However, the shorter the list of recommended plans, the more recommended income may be lost. For example, suppose a game equipment enhancement material recommendation activity includes plan X, plan Y, and plan Z, among which, plan X, plan Y, and plan Z represent common enhancement materials, rare enhancement materials, and epic enhancement materials respectively, and the preset ratios are 25%, 50%, and 25% respectively. Set the game players into two groups, one is wealthy players and the other is ordinary players. For wealthy players, they will buy whatever is recommended. Obviously, recommending epic enhancement materials (plan Z) can bring the highest income. For ordinary players, they can only afford common enhancement materials (plan X) and rare enhancement materials (plan Y). Based on this, the biggest profit plan is to leave 25% of the opportunities to recommend epic enhancement materials (plan Z) to wealthy players. However, in actual game operation, the login and purchase behaviors of rich and ordinary players are not continuous and uniform. It is very likely that in a certain short period of time, many rich players are active, and in a certain short period of time, few rich players are active. If the recommendation plan list is very short, for example: there are only 4 candidate recommendation plans, that is, ordinary enhancement materials (plan X), rare enhancement materials (plan Y), rare enhancement materials (plan Y), and epic enhancement materials (plan Z), then, if 4 rich players in a row trigger recommendation requests, only 1 rich player can be recommended epic reinforcement materials (plan Z), and the purchasing power of the remaining 3 wealthy players will be wasted. But if the recommendation plan list is relatively long, for example: there are 4,000 candidate recommendation plans, among which there are 1,000 epic enhancement materials (plan Z), then, unless rich players trigger recommendation requests more than 1,000 times among the 4,000 recommendation requests, it will not cause a waste of purchasing power of rich players. The longer the queue, the less gain you lose.

Therefore, in the design stage of the recommendation activity, it is necessary to rationalize the length of the recommendation plan list, so as to balance the deviation, balance time, and recommendation benefits among the preset recommendation plans.

The second embodiment of the present application provides a recommendation server. FIG. 7 is a schematic structural diagram of a recommendation server provided in this embodiment.

As shown in FIG. 7 , the recommendation server provided in this embodiment includes: a recommended solution list construction unit 701, a recommended solution acquisition unit 702, a candidate recommended solution query unit 703, and a recommendation unit 704;

The recommended solution list construction unit 701 is configured to construct a recommended solution list, the recommended solution list includes N candidate recommended solutions, and among the N candidate recommended solutions, preset recommended solutions of M categories satisfy a preset quantity ratio, wherein N and M are both positive integers greater than 1.

Optionally, the construction of the list of recommended solutions includes:

Acquiring preset recommendation schemes of the M categories;

The number N of the candidate recommendation solutions is determined according to the preset balance time of the preset recommendation solutions for the M categories and the number of acquisitions of the recommendation requests within the preset balance time, wherein the preset balance time is the unit duration representing the consumption of the candidate recommendation solutions in the recommendation solution list;

Determine the amplification factor according to the number N of the candidate recommendation schemes and the preset quantity ratio;

The preset recommendation scheme of each category is amplified according to the amplification factor to form the recommendation scheme list.

The recommendation scheme acquiring unit 702 is configured to acquire a first recommendation scheme for the user in response to receiving a recommendation request from the user.

Optionally, the recommendation request includes user characteristic information of the user; and obtaining the first recommendation plan for the user in response to receiving the user's recommendation request includes:

inputting the recommendation request of the user into a preset recommendation model, so that the recommendation model determines a plurality of recommendation schemes for the user according to the user characteristic information in the recommendation request;

sorting the plurality of recommendation schemes according to the expected virtual value from high to low;

Outputting the first recommendation solution from the plurality of recommendation solutions sorted from high to low by the expected virtual value.

Optionally, the first recommendation scheme is the recommendation scheme with the highest expected virtual value among the multiple recommendation schemes; and outputting the first recommendation scheme from the multiple recommendation schemes sorted from high to low by the expected virtual value includes:

Selecting the recommendation solution with the highest expected virtual value from the plurality of recommendation solutions is output as the first recommendation solution.

The candidate recommendation solution query unit 703 is configured to search the recommendation solution list for a first candidate recommendation solution of the same category as the first recommendation solution.

The recommending unit 704 is configured to, in response to finding the first candidate recommendation solution in the recommendation solution list, recommend the first candidate recommendation solution to the user, and delete the first candidate recommendation solution from the recommendation solution list.

Optionally, the recommendation server is also used for:

In response to not finding the first candidate recommendation solution in the list of recommendation solutions, acquiring a second recommendation solution for the user, wherein the second recommendation solution is of a different category from the first recommendation solution;

Querying the recommended second proposal list for a second candidate recommended proposal of the same category as the second recommended proposal;

In response to finding the second candidate recommendation solution in the recommendation solution list, recommending the second candidate recommendation solution to the user, and deleting the second candidate recommendation solution from the recommendation solution list.

Optionally, the recommendation server is also used for:

In response to not finding the first candidate recommendation solution in the list of recommendation solutions, and not finding any of the candidate recommendation solutions in the list of recommendation solutions, generating a second recommendation solution list, wherein the preset recommendation solutions of the M categories in the second recommendation solution list satisfy a preset quantity ratio;

Querying the third candidate recommendation plan in the second recommendation plan list for the same category as the first recommendation plan;

In response to finding the third candidate recommendation solution in the second recommendation solution list, recommending the third candidate recommendation solution to the user, and deleting the third candidate recommendation solution from the second recommendation solution list.

The third embodiment of the present application provides a recommendation system. FIG. 8 is a schematic structural diagram of the recommendation system provided by this embodiment.

As shown in FIG. 8 , the recommendation system provided by this embodiment includes: a user terminal 801 , a service server 802 , and a recommendation server 803 as described in the second embodiment of the present application.

The client 801 includes: a first receiving module 8011, a first processing module 8012, and a first sending module 8013;

The first receiving module 8011 is configured to receive an operation instruction from a user, and the operation is performed as an action that can trigger a recommended behavior;

The first processing module 8012 is configured to generate instruction information corresponding to the operation instruction according to the operation instruction;

The first sending module 8013 is configured to send the instruction information to the service server 802;

The business server 802 includes: a second receiving module 8021, a second processing module 8022, and a second sending module 8023;

The second receiving module 8021 is configured to receive the instruction information sent by the client 801;

The second processing module 8022 is configured to generate a recommendation request for the user according to the instruction information;

The second sending module 8023 is configured to send the recommendation request to the recommendation server 803;

The recommendation server 803 includes: a third receiving module 8031;

The third receiving module 8031 is configured to receive the recommendation request sent by the business server 802;

The recommendation server 803 also includes a recommended solution list construction unit 8032, a recommended solution acquisition unit 8033, a candidate recommended solution query unit 8034, and a recommended solution unit 8035 as described in the second embodiment of the present application;

The recommendation plan list construction unit 8032 is configured to construct a recommendation plan list, the recommendation plan list includes N candidate recommendation plans, and the preset recommendation plans of M categories in the N candidate recommendation plans satisfy a preset quantity ratio, wherein N and M are both positive integers greater than 1;

The recommendation scheme acquiring unit 8033 is configured to acquire a first recommendation scheme for the user in response to receiving the recommendation request;

The candidate recommendation solution query unit 8034, configured to query the first candidate recommendation solution of the same category as the first recommendation solution in the recommendation solution list;

The recommending unit 8035 is configured to, in response to finding the first candidate recommendation solution in the recommendation solution list, recommend the first candidate recommendation solution to the user, and delete the first candidate recommendation solution from the recommendation solution list.

The fourth embodiment of the present application provides an electronic device. FIG. 9 is a schematic structural diagram of the electronic device provided by this embodiment.

As shown in FIG. 9 , the electronic device provided in this embodiment includes: a memory 901 and a processor 902 .

The memory 901 is used for storing computer instructions for executing the recommendation method.

The processor 902 is configured to execute computer instructions stored in the memory 901, and perform the following operations:

Constructing a list of recommendation schemes, the list of recommendation schemes includes N candidate recommendation schemes, and the preset recommendation schemes of M categories in the N candidate recommendation schemes satisfy a preset quantity ratio, wherein N and M are both positive integers greater than 1;

In response to receiving a recommendation request from a user, obtain a first recommendation solution for the user;

querying the list of recommended solutions for a first candidate recommended solution of the same category as the first recommended solution;

In response to finding the first candidate recommendation solution in the recommendation solution list, recommending the first candidate recommendation solution to the user, and deleting the first candidate recommendation solution from the recommendation solution list.

Optionally, the construction of the list of recommended solutions includes:

Acquiring preset recommendation schemes of the M categories;

The number N of the candidate recommendation solutions is determined according to the preset balance time of the preset recommendation solutions for the M categories and the number of acquisitions of the recommendation requests within the preset balance time, wherein the preset balance time is the unit duration representing the consumption of the candidate recommendation solutions in the recommendation solution list;

Determine the amplification factor according to the number N of the candidate recommendation schemes and the preset quantity ratio;

The preset recommendation scheme of each category is amplified according to the amplification factor to form the recommendation scheme list.

Optionally, the recommendation request includes user characteristic information of the user; and obtaining the first recommendation plan for the user in response to receiving the user's recommendation request includes:

inputting the recommendation request of the user into a preset recommendation model, so that the recommendation model determines a plurality of recommendation schemes for the user according to the user characteristic information in the recommendation request;

sorting the plurality of recommendation schemes according to the expected virtual value from high to low;

Outputting the first recommendation solution from the plurality of recommendation solutions sorted from high to low by the expected virtual value.

Optionally, the first recommendation scheme is the recommendation scheme with the highest expected virtual value among the multiple recommendation schemes; and outputting the first recommendation scheme from the multiple recommendation schemes sorted from high to low by the expected virtual value includes:

Selecting the recommendation solution with the highest expected virtual value from the plurality of recommendation solutions is output as the first recommendation solution.

Optionally, it is also used to perform the following operations:

In response to not finding the first candidate recommendation solution in the list of recommendation solutions, acquiring a second recommendation solution for the user, wherein the second recommendation solution is of a different category from the first recommendation solution;

Querying the recommended second proposal list for a second candidate recommended proposal of the same category as the second recommended proposal;

In response to finding the second candidate recommendation solution in the recommendation solution list, recommending the second candidate recommendation solution to the user, and deleting the second candidate recommendation solution from the recommendation solution list.

Optionally, it is also used to perform the following operations:

In response to not finding the first candidate recommendation solution in the list of recommendation solutions, and not finding any of the candidate recommendation solutions in the list of recommendation solutions, generating a second recommendation solution list, wherein the preset recommendation solutions of the M categories in the second recommendation solution list satisfy a preset quantity ratio;

Querying the third candidate recommendation plan in the second recommendation plan list for the same category as the first recommendation plan;

In response to finding the third candidate recommendation solution in the second recommendation solution list, recommending the third candidate recommendation solution to the user, and deleting the third candidate recommendation solution from the second recommendation solution list.

The fifth embodiment of the present application provides a computer-readable storage medium, the computer-readable storage medium includes computer instructions, and the computer instructions are used to implement the methods described in the various embodiments of the present application when executed by a processor.

It should be noted that the relational terms such as "first" and "second" herein are only used to distinguish one entity or operation from another entity or operation, and do not require or imply any actual relationship or order between these entities or operations. In addition, "including", "has", "comprising" and "including" and other similar forms of words have the same meaning, and the end of any one or more items after any of the above words is open-ended, and any of the above nouns does not mean that the one or more items have been listed exhaustively, or are limited to these listed one or more items.

As used herein, unless expressly stated otherwise, the term "or" includes all possible combinations except where infeasible. For example, if it is expressed that a database may include A or B, then unless specifically specified or impracticable otherwise, database A, or B, or both A and B may be included. As a second example, if it is expressed that a certain database may include A, B, or C, then unless otherwise specified or impracticable, the database may include database A, or B, or C, or A and B, or A and C, or B and C, or A and B and C.

It should be noted that the above-described embodiments can be realized by hardware or software (program code), or a combination of hardware and software. If implemented by software, it can be stored on the aforementioned computer-readable medium. When the software is executed by the processor, it can execute the above-mentioned disclosed method. The computing unit and other functional units described in this disclosure may be realized by hardware or software, or a combination of hardware and software. Those of ordinary skill in the art will also understand that the above-mentioned multiple modules/units can be combined into one module/unit, and each of the above-mentioned modules/units can be further divided into multiple sub-modules/sub-units.

In the foregoing detailed description, embodiments have been described with reference to numerous specific details that may vary from implementation to implementation. Certain adaptations and modifications may be made to the described embodiments. Those skilled in the art can obviously obtain some other implementation manners from the specific implementation manners disclosed in this application. The specification and examples are intended for purposes of illustration only, with the true scope and nature of the application indicated by the appended claims. The order of the steps shown in the figure is only for the purpose of explanation, and does not mean to be limited to any specific steps and order. Therefore, those skilled in the art will realize that when performing the same method, these steps may be performed in a different order.

In the drawings and detailed description of this application, exemplary embodiments are disclosed. However, many variations and modifications can be made to these embodiments. Accordingly, although specific terms are used, these terms are general and descriptive only and not for purposes of limitation.

Claims (10)

1. A recommendation method, the method comprising:

constructing a recommendation scheme list, wherein the recommendation scheme list comprises N candidate recommendation schemes, and preset quantity ratios among M categories of preset recommendation schemes in the N candidate recommendation schemes are met, wherein N and M are positive integers larger than 1;

Responding to a received recommendation request of a user, and acquiring a first recommendation scheme aiming at the user;

querying a first candidate recommended scheme which is the same as the first recommended scheme category in the recommended scheme list;

and recommending the first candidate recommended scheme to the user in response to the first candidate recommended scheme being queried in the recommended scheme list, and deleting the first candidate recommended scheme from the recommended scheme list.

2. The method according to claim 1, wherein the method further comprises:

obtaining a second recommendation for the user in response to the first candidate recommendation not being queried in the recommendation list, wherein the second recommendation is different from the first recommendation in category;

querying a second candidate recommended solution which is the same as the second recommended solution category in the recommended solution list;

and recommending the second candidate recommended scheme to the user in response to the second candidate recommended scheme being queried in the recommended scheme list, and deleting the second candidate recommended scheme from the recommended scheme list.

3. The method according to claim 1, wherein the method further comprises:

Generating a second recommendation scheme list in response to the fact that the first candidate recommendation scheme is not queried in the recommendation scheme list and any one candidate recommendation scheme is not queried in the recommendation scheme list, wherein preset quantity ratios among M categories of preset recommendation schemes in the second recommendation scheme list are met;

querying a third candidate recommended solution which is the same as the first recommended solution category in the second recommended solution list;

and recommending the third candidate recommended solution to the user in response to the inquiry of the third candidate recommended solution in the second recommended solution list, and deleting the third candidate recommended solution from the second recommended solution list.

4. The method of claim 1, wherein the building a list of recommendations comprises:

acquiring preset recommendation schemes of the M categories;

determining the number N of the candidate recommended schemes according to preset balance time of the preset recommended schemes aiming at the M categories and the acquisition times of the recommended requests in the preset balance time, wherein the preset balance time is unit duration for representing that the candidate recommended schemes in the recommended scheme list are consumed;

Determining amplification factors according to the number N of the candidate recommended schemes and the preset number proportion;

amplifying the preset recommended schemes of each category according to the amplification multiples to form the recommended scheme list.

5. The method of claim 1, wherein the recommendation request includes user characteristic information of the user; the responding to the received recommendation request of the user, the first recommendation scheme aiming at the user is obtained, and the method comprises the following steps:

inputting the recommendation request of the user into a preset recommendation model, so that the recommendation model determines a plurality of recommendation schemes for the user according to the user characteristic information in the recommendation request;

ranking the plurality of recommendations from high to low according to expected virtual value;

outputting the first recommended scheme from the plurality of recommended schemes with the expected virtual value ranked from high to low.

6. The method of claim 5, wherein the first recommendation is the recommendation having the highest expected virtual value of the plurality of recommendations; the outputting the first recommendation scheme from the plurality of recommendation schemes after the expected virtual value is ranked from high to low comprises the following steps:

And selecting the recommended scheme with the highest expected virtual value from the plurality of recommended schemes to output as the first recommended scheme.

7. The recommendation server is characterized in that the recommendation server comprises: the recommendation system comprises a recommendation scheme list construction unit, a recommendation scheme acquisition unit, a candidate recommendation scheme query unit and a recommendation unit;

the recommendation scheme list construction unit is used for constructing a recommendation scheme list, wherein the recommendation scheme list comprises N candidate recommendation schemes, and preset quantity proportions of M categories of preset recommendation schemes in the N candidate recommendation schemes are met, wherein N and M are positive integers which are larger than 1;

the recommendation scheme obtaining unit is used for obtaining a first recommendation scheme aiming at a user in response to receiving a recommendation request of the user;

the candidate recommended solution inquiring unit is used for inquiring a first candidate recommended solution with the same category as the first recommended solution in the recommended solution list;

the recommending unit is used for recommending the first candidate recommended scheme to the user in response to the fact that the first candidate recommended scheme is inquired in the recommended scheme list, and deleting the first candidate recommended scheme from the recommended scheme list.

8. A recommendation system, the system comprising: a user terminal, a service server terminal, and a recommendation server terminal as described in claim 7; wherein,,

the user terminal comprises: the device comprises a first receiving module, a first processing module and a first sending module;

the first receiving module is used for receiving an operation instruction of a user, wherein the operation instruction is an action capable of triggering a recommendation behavior;

the first processing module is used for generating instruction information corresponding to the operation instruction according to the operation instruction;

the first sending module is used for sending the instruction information to the service server;

the business server side comprises: the device comprises a second receiving module, a second processing module and a second sending module;

the second receiving module is used for receiving the instruction information sent by the user side;

the second processing module is used for generating a recommendation request for the user according to the instruction information;

the second sending module is used for sending the recommendation request to the recommendation server;

the recommendation server side comprises: a third receiving module;

the third receiving module is configured to receive the recommendation request sent by the service server.

9. An electronic device, comprising: a memory, a processor;

the memory is used for storing one or more computer instructions;

the processor being configured to execute the one or more computer instructions to implement the method of any of claims 1-6.

10. A computer readable storage medium having stored thereon one or more computer instructions which, when executed by a processor, perform the method of any of claims 1-6.