CN106997347A - Information recommendation method and server - Google Patents

Abstract

Embodiments of the present invention provide an information recommendation method and server. The information recommendation method of the present invention may include: grouping a plurality of users to obtain at least two user groups; each user group includes at least one user; determining a recommendation strategy corresponding to each user group; the recommendation strategy includes each user group corresponding A recommendation algorithm; generating recommendation information for each user group according to a recommendation strategy corresponding to each user group; sending the recommendation information to a terminal corresponding to at least one user. The embodiments of the present invention can improve the accuracy of information recommendation.

Description

Information recommendation method and server

technical field

Embodiments of the present invention relate to communication technologies, and in particular to an information recommendation method and server.

Background technique

The emergence and popularization of Internet technology has brought a large amount of information to users and met the needs of users for information in the information age. However, with the rapid development of the network, the information has increased significantly, making users unable to deal with a large amount of information. If you find some information that is useful to you, the efficiency of using the information will be reduced, resulting in the problem of information overload. Information recommendation technology is created to solve this problem.

The commonly used information recommendation technology is actually to determine the recommendation algorithm for the large user group, and determine the recommended information for the large user group according to the recommendation algorithm, and then send the recommended information to the large user group.

Because in this information recommendation technology, the same recommendation information is determined for a large user group according to the recommendation algorithm, the recommendation information cannot reflect the user's personality, and the accuracy of information recommendation is low.

Contents of the invention

Embodiments of the present invention provide an information recommendation method and server to improve the accuracy of information recommendation.

In the first aspect, an information recommendation method is provided, including:

Grouping multiple users to obtain at least two user groups; wherein, each user group includes at least one user;

Determine a recommendation strategy corresponding to each user group; wherein, the recommendation strategy may include a recommendation algorithm corresponding to each user group;

Generate recommendation information for each user group according to the recommendation strategy corresponding to each user group;

Send recommendation information to a terminal corresponding to at least one user.

Optionally, multiple users are grouped to obtain at least two user groups, which may include:

According to the data feature of each of the multiple users, the multiple users are grouped to obtain the at least two user groups.

Optionally, the data characteristics of each user include: at least one data characteristic;

According to the data characteristics of each of the multiple users, group the multiple users to obtain at least two user groups, which may include:

According to at least one data feature of each user, multiple users are grouped according to a preset sequence of data features to obtain at least two user groups.

The information recommendation algorithm provided by the embodiment of the present invention can group the multiple users according to the preset order of data characteristics according to at least one data characteristic of the users, which can make the grouping of users more accurate, so that each user group can The recommendation algorithm adopted by all users is the same, which improves the accuracy of information recommendation.

Optionally, the data characteristics of each user include at least one of the following:

Per-user history data, per-user tag data, per-user attribute data, and per-user social data.

Wherein, the historical record data of each user includes: at least one of all historical record data of each user and historical record data of each user within a preset time period; the tag data of each user includes: each user At least one of the number of effective tags and the number of prominent tags for each user; the attribute data of each user includes at least one of the gender of each user, the age group of each user, and the identity of each user; The social data of a user includes at least one of the number of contacts of each user, the number of close contacts of each user, and the number of active contacts of each user; the close contacts of each user are the number of contacts with each user Contacts greater than or equal to the preset threshold.

Optionally, determine the recommendation strategy corresponding to each user group, including:

Determine the characteristic value range to which the data characteristic value of each user group belongs;

The recommendation algorithm corresponding to the feature value range is used as the recommendation algorithm corresponding to each user group.

Optionally, before determining the characteristic value range to which the data characteristic value of each user group belongs, the method may further include:

Determine data feature values for each user group.

Optionally, the data characteristics of each user group include at least one of the following: historical record data of each user group, label data of each user group, attribute data of each user group, and social data of each user group; The data characteristic value of each user group includes: a characteristic value corresponding to each data characteristic of each user group.

Optionally, using the recommendation algorithm corresponding to the feature value range as the recommendation algorithm corresponding to each user group includes:

According to the feature value range to which the data feature value of each user group belongs, the recommendation algorithm corresponding to the feature value range in the preset recommendation algorithm library is determined, and the determined recommendation algorithm is used as the recommendation algorithm corresponding to each user group.

Optionally, if the recommendation algorithm corresponding to each user group includes multiple recommendation algorithms; the recommendation strategy includes: N fusion recommendation algorithms corresponding to each user group; wherein, each fusion recommendation algorithm includes multiple recommendation algorithms and multiple A fusion method corresponding to the recommendation algorithm;

Identify recommended strategies for each user group, which may include:

According to the recommendation algorithm corresponding to each user group, and, the usage conditions of each fusion method in the preset fusion method library for the recommendation algorithm, determine N from the recommendation algorithm library that meets the usage conditions of the recommendation algorithm corresponding to each user group The fusion methods are N fusion methods corresponding to multiple recommendation algorithms;

According to the N fusion methods, multiple recommendation algorithms are respectively fused to obtain N fusion recommendation algorithms.

Optionally, generate recommendation information for each user group according to the recommendation strategy corresponding to each user group, including:

Determine the recommendation accuracy of each fusion recommendation algorithm corresponding to each user group;

Determine the fusion recommendation algorithm with the highest recommendation accuracy from N fusion recommendation algorithms as the optimal fusion recommendation algorithm for each user group;

According to the optimal fusion recommendation algorithm of each user group, the recommendation information of each user group is generated.

The information recommendation algorithm provided by the embodiment of the present invention, by determining the optimal fusion recommendation algorithm for each user group, and then generating the recommendation information for each user group according to the optimal fusion recommendation algorithm, can make the information recommendation for each user group Maximize the accuracy and ensure the accuracy of information recommendation.

In a second aspect, a server is provided, including a module for executing the method in the first aspect.

In a third aspect, a server is provided, including a processor, a memory, a communication interface and a bus, and the processor is connected to the memory and the communication interface through the bus; the memory is used to store instructions; the processor is used to execute the instructions; when the processor executes the memory storage instructions, causing the processor to execute the method described in the first aspect.

In a fourth aspect, a computer-readable storage medium is provided, in which executable program code is stored, and the program code is used to implement the method described in the first aspect.

The information recommendation method and server provided by the embodiments of the present invention can obtain at least two user groups by grouping multiple users, determine the recommendation strategy corresponding to each user group, and generate the recommendation information of each user group, and send the recommendation information to a terminal corresponding to at least one user in each user group. Because the information recommendation method can determine the recommendation strategy for each divided user group, and then determine the recommendation information of each user group, it can reflect the user personality among different user groups and ensure the accuracy of information recommendation.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description These are some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained according to these drawings without any creative effort.

FIG. 1 is an architecture diagram of an information recommendation system applicable to an information recommendation method provided by an embodiment of the present invention;

FIG. 2 is a schematic structural diagram of a computer corresponding to a server provided by an embodiment of the present invention;

FIG. 3 is a flowchart of an information recommendation method provided by an embodiment of the present invention;

FIG. 4 is a flowchart of another information recommendation method provided by an embodiment of the present invention;

Fig. 5 is a correspondence diagram of a user group, fusion recommendation strategy, data group and recommendation accuracy of recommendation information provided by an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a server provided by an embodiment of the present invention;

FIG. 7 is a schematic structural diagram of a computer-readable storage medium provided by an embodiment of the present invention.

detailed description

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

The information recommendation method provided by each embodiment of the present invention is applicable to information recommendation systems in the field of Internet technology, such as e-commerce or social software. FIG. 1 is a structural diagram of an information recommendation system to which an information recommendation method provided by an embodiment of the present invention is applicable. As shown in FIG. 1 , the information recommendation system may include a server 101 and at least one terminal 102 . The server 101 may be connected to the at least one terminal 102 through a wired network or a wireless network. The server 101 may execute the information recommendation method provided by the embodiment of the present invention to recommend corresponding information for each terminal 102 . The server 101 may be an independent server, or a server cluster composed of several servers, or a server in a data center. The server 101 may be, for example, a service server in an information recommendation system. The terminal 102 can present the information recommended by the server to the user, and the terminal 101 can also perform corresponding operations on the recommended information according to user instructions, such as responding or not responding. The terminal 101 can be, for example, a smart phone, a tablet computer, an e-reader, a notebook computer or a desktop computer, and the like.

FIG. 2 is a schematic structural diagram of a computer corresponding to a server provided by an embodiment of the present invention. As shown in FIG. 2 , the server 200 may include: a processor 201 , a memory 202 , a network interface 203 and a communication bus 204 . The processor 201 , memory 202 and network interface 203 are connected and communicated through a communication bus 204 .

Wherein, the processor 201 may be a central processing unit (Central Processing Unit, CPU for short). The processor 201 may also be other general-purpose processors, digital signal processors (Digital Signal Processing, DSP for short), application specific integrated circuits (Application Specific Integrated Circuit, ASIC for short), field-programmable gate arrays (Field-Programmable Gate Array, FPGA for short), ) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like.

The memory 202 may include a volatile memory (volatile memory), such as a random access memory (Random-Access Memory, RAM for short); the memory 202 may also include a non-volatile memory (Non-volatile memory), such as a read-only memory (Read-Only Memory, be called for short ROM), flash memory (Flash Memory), hard disk (Hard Disk Drive, be called for short HDD) or solid state drive (Solid-State Drive, be called for short SSD); Memory 202 can also comprise the memory of above-mentioned kind combination.

The network interface 203 can be an interface circuit for sending and receiving information. After the network interface 203 receives the information sent by the external device, it can be processed by the processor 201; in addition, the network interface 203 can communicate with the network and other devices through wired communication or wireless communication.

The server 200 in FIG. 2 may be the server 101 in the above-mentioned information recommendation system in FIG. 1 . The server 200 may determine the recommendation information through the processor 201, and send the recommendation information to a terminal corresponding to the user through the network interface 203.

Specifically, the processor 201 may execute the information recommendation method provided by each embodiment of the present invention by calling the program or instruction stored in the memory 202, thereby obtaining the recommendation information of each user group, and passing the recommendation information of each user group through The communication bus 204 transmits to the network interface 203 . The network interface 203 may send the recommendation information of each user group to terminals corresponding to the users of each user group.

FIG. 3 is a flow chart of an information recommendation method provided by an embodiment of the present invention. The method is performed by the server 200 as described above. As shown in Figure 3, the method may include:

S301. Group a plurality of users to obtain at least two user groups; where each user group includes at least one user.

Specifically, the multiple users may be multiple users in a large user group. The large user group may be a user group that has not been divided into types, and the large user group may be, for example, a user group with more than 1,000 users. The server may use a clustering algorithm (Clustering Algorithm) to group the multiple users according to the data of the multiple users according to the number of groups. Wherein, the similarity of user data in the same user group is within a preset range. That is to say, the data similarity of users in the same user group is high, and the data similarity of users of different user groups is low. The clustering algorithm may be, for example, a K-means clustering algorithm. The number of groups, that is, the number of user groups after grouping, may be a preset number of groups, or may be an optimal number of groups determined by the server using Bayesian information criterion according to the interval of the preset number of groups.

S302. Determine a recommendation strategy corresponding to each user group, where the recommendation strategy includes: a recommendation algorithm corresponding to each user group.

S303. Generate recommendation information for each user group according to the recommendation policy corresponding to each user group.

Specifically, the server may determine the recommendation policy corresponding to each user group according to the data of the users in each user group. The same user group corresponds to the same recommendation strategy, and different user groups correspond to different recommendation strategies. The recommendation strategy is used to generate recommendation information.

Wherein, the recommendation algorithm corresponding to each user group may be selected from a preset recommendation algorithm library. The preset recommendation algorithm library can include content-based (Content-Based) recommendation algorithms, user-based (User-Based) collaborative filtering recommendation algorithms, item-based (Item-Based) collaborative filtering recommendation algorithms, tag-based recommendation algorithms At least one recommendation algorithm among algorithm, collaborative filtering recommendation algorithm based on matrix decomposition, recommendation algorithm based on user attributes, and recommendation algorithm based on social interaction.

For example, the content-based recommendation algorithm may determine recommendation information according to the service content that the user is interested in before the current time. If the recommendation strategy corresponding to each user group includes: a content-based recommendation algorithm, then the recommendation information for each user group may include: similarity to the business content that each user in each user group is interested in before the current time Higher business content. The service content may include, for example: the user's favorite service item (Item) represented by the user in the form of text, picture or video.

The user-based collaborative filtering recommendation algorithm can, for example, determine the similarity between the user and other users according to the business content that multiple users are interested in before the current time, and determine the similarity between the user and other users with high similarity to the user. The business content of interest is determined as the recommendation information of the user. For example, if the business content that user A is interested in includes item A and item C, the business content that user B is interested in includes item B, and the business content that user C is interested in includes item A, item C, and item D, then user C The similarity with user A is high, and the recommendation information of user A may include service content, such as item D, that user C is interested in. If the recommendation strategy for each user group includes: a user-based collaborative filtering recommendation algorithm, then the recommendation information for each user group may include information that is of interest to other users with a higher similarity with each user in each user group business content. The service content may include, for example: the user's favorite service items represented by the user in the form of text, pictures or videos, and the similarity between the user and other users may be the similarity between the user and the other users' preferences.

The item-based collaborative filtering recommendation can be, for example, according to the business content that multiple users are interested in before the current time, determine the business content with high similarity with the business content that the user is interested in, and use the high similarity The service content is determined as the user's recommended information. For example, if the business content that user A is interested in includes item A, the business content that user B is interested in includes item A, item B, and item C, and the business content that user C is interested in includes item A and item C, then the same as item The service content with a high similarity to A is item C, and the recommendation information of the user A may include the item C. If the recommendation strategy for each user group includes: an item-based collaborative filtering recommendation algorithm, then the recommendation information for each user group may include: information with a high similarity to the business content that each user in each user group is interested in Other business content.

The label-based recommendation algorithm may, for example, determine the service content label that the user is interested in according to the service content that the user is interested in before the current time, and determine the service content with the service content label as the user's recommendation information. The service content tag may be, for example, a textual expression tag, such as "romantic", "80s" and so on. If the recommendation strategy for each user group includes: a tag-based recommendation algorithm, then the recommendation information for each user group may include: service content with a service content tag of each user in each user group.

The recommendation algorithm based on matrix decomposition may, for example, be to add matrix decomposition technology to the above-mentioned user-based or item-based collaborative filtering recommendation algorithm, and decompose the matrix composed of the business content that the multiple users are interested in before the current time, Obtaining user information and business content information, and then determining user recommendation information based on user information and/or business content information can solve the problem of data sparsity.

The recommendation algorithm based on user attributes may, for example, determine all the service content that the user is interested in corresponding to the user attributes as the user's recommendation information. If the recommendation strategy for each user group includes: a recommendation algorithm based on user attributes, then the recommendation information for each user group may include: business content of interest to all users corresponding to the attributes of each user in each user group .

The social-based recommendation algorithm may, for example, determine the service content of interest to all users corresponding to the social situation as the user's recommendation information according to the user's social situation. If the recommendation strategy for each user group includes: a social-based recommendation algorithm, then the recommendation information for each user group may include: business content of interest to all users corresponding to the social status of each user in each user group .

S304. Send the recommendation information to a terminal corresponding to the at least one user.

Specifically, the server may send the recommendation information to a terminal corresponding to the at least one user through a network interface.

The information recommendation method provided by the embodiment of the present invention can obtain at least two user groups by grouping a plurality of users, determine the recommendation strategy corresponding to each user group, and generate each user group according to the recommendation strategy corresponding to each user group. recommend information for each user group, and send the recommend information to the terminal corresponding to at least one user in each user group. Because the information recommendation method can determine the recommendation strategy for each divided user group, and then determine the recommendation information of each user group, it can reflect the user personality among different user groups and ensure the accuracy of information recommendation.

Optionally, grouping multiple users in S301 as described in the above embodiment to obtain at least two user groups may include:

According to the data feature of each of the multiple users, the multiple users are grouped to obtain the at least two user groups.

Optionally, the data feature of each user may include: at least one data feature.

In the steps described above, according to the data characteristics of each of the multiple users, the multiple users are grouped, and obtaining the at least two user groups may include:

According to the at least one data feature of each user, the multiple users are grouped according to a preset sequence of data features to obtain the at least two user groups.

Specifically, the server may, according to the at least one data feature of each user, group the multiple users level by level according to the preset sequence of data features, and then obtain the at least two user groups. Wherein, the grouping of the multiple users according to the at least one data feature may be performed by using a clustering algorithm; the clustering algorithms used for grouping according to different data features may be the same or different.

For example, if the data features of each user include 3 data features; the preset sequence of data features may be data feature 1, data feature 2, and data feature 3. Then, the server may, according to the preset sequence of data characteristics, first classify the plurality of users according to data characteristic 1, and then perform secondary grouping on the first-level grouped user groups according to the data characteristic 2, and According to the data characteristic 3, the user group after the second-level grouping is grouped into a third-level group, and then the at least two user groups are obtained.

In the following, the server performs one-level grouping according to a data feature sampling clustering algorithm as an example for illustration. If the one data characteristic includes: various statistical data, the server may use a clustering algorithm such as the K-means algorithm to group the multiple users according to the six statistical data, thereby dividing the multiple users into multiple user groups. Specifically, the server uses the K-means algorithm to group according to the six statistical data, and the specific process can be as follows:

(1) Randomly select the data features of K users as the centroid, and K is the preset number of groups;

(2) Measure the distance from each remaining user to each centroid, and classify each user into the user group corresponding to the nearest centroid;

(3) recalculate the centroid of each user group that has been obtained;

(4) Iterate 2-3 steps until the new centroid is equal to the original centroid or less than the preset threshold.

Optionally, the data characteristics of each user may include at least one of the following: historical record data of each user, label data of each user, attribute data of each user, and social data of each user .

Specifically, according to the at least one data characteristic of each user in the above-mentioned steps, the multiple users are grouped according to the preset order of data characteristics, and the obtaining of the at least two user groups may include: according to the user All historical record data of multiple users are clustered at the first level, and the user groups after the first-level clustering are clustered at the second level based on the recent historical record data of the users. The user groups after the second-level clustering are clustered according to Three-level clustering is performed on the user's label data, four-level clustering is performed on the user group after the three-level clustering according to the user's attribute data, and the last clustering is performed on the user group after the four-level clustering based on the user's social data. Clustering to get at least two user groups.

Wherein, the historical record data of each user includes: at least one of all historical record data of each user and historical record data of each user within a preset time period.

The tag data of each user includes: at least one of the number of valid interest tags for each user and the number of prominent interest tags for each user.

The attribute data of each user includes at least one of the gender of each user, the age group of each user, and the identity of each user.

The social data of each user includes at least one of the number of contacts of each user, the number of close contacts of each user, and the number of active contacts of each user; the close contacts of each user are The contacts whose contact times of each user are greater than or equal to the preset times threshold.

Specifically, all historical record data of each user may be, for example, all historical record data of each user from the service registration time to the current time. The historical record data of each user in the preset period of time may be, for example, the historical record data of each user in the preset period closest to the current time, such as the week or week closest to the current time for each user. Historical data for one month. For example, if the information recommendation method is applied to an information recommendation system in the field of e-commerce, then all historical record data of each user may include: browsing history record data of each user, purchase history of each user At least one of record data, each user's favorite history record data, and the like. Wherein, the browsing history data of each user includes historical record data of items browsed by each user, or historical record data of entries of items browsed by each user. The purchase history data of each user includes: historical record data of items purchased by each user, or historical record data of items purchased by each user. The collection history record data of each user includes: the history record data of each user's collection items, or the history record data of the entries of each user's collection items. Wherein, the historical record data of items browsed by each user is different from the historical record data of items browsed by each user. The historical record data of each user's favorite items is different from the historical record data of each user's favorite item entries. The historical record data of items purchased by each user is different from the historical record data of items purchased by each user. Taking the purchase of items as an example, if a user purchases the same item multiple times, the total number of purchase records of the user is greater than the number of items purchased.

The number of effective interest tags for each user can be the number of interest tags for each user within the preset validity period; the number of prominent interest tags for each user can be the tag weight of each user's effective interest tags The number of labels greater than the preset weight value. Each tag can have a tag weight, which can be used to indicate the importance of each tag to the user. The greater the weight of the label, the greater the importance of each label to the user. The label weight of each label can be adjusted according to the change of time. For example, if the type tag of a movie watched by the user was generated a month ago, and the user no longer watches this type of movie, the weight should be lowered.

The gender of each user may include male or female. The age group of each user may include, for example: any one of children, young people, middle-aged people and old people. The identity of each user may include: any one of a student, a housewife, and an office worker.

The number of contacts of each user may be, for example, the total number of contacts in each user's address book or social software. The close contact of each user is a contact in each user's mobile phone address book or social software whose number of contacts with the user is greater than or equal to a preset number of times threshold. The number of contacts of each user, the number of close contacts of each user, and the number of active contacts of each user may be obtained from communication records or social records of each user.

FIG. 4 is a flowchart of another information recommendation method provided by an embodiment of the present invention. As shown in FIG. 4, the method is based on the methods described in the above-mentioned embodiments, wherein determining the recommendation strategy corresponding to each user group in S302 may include:

S401. Determine the characteristic value range to which the data characteristic value of each user group belongs.

S402. Use the recommendation algorithm corresponding to the feature value range as the recommendation algorithm corresponding to each user group.

Specifically, if the data feature of each user in each user group includes at least one data feature, then the data feature value of each user group includes: a feature value corresponding to at least one data feature. Then in S401, it may be to determine the feature value range to which the data feature value corresponding to each data feature in the at least one data feature belongs, and in S402, it may be to determine the data feature value corresponding to each data feature in the at least one data feature. The recommendation algorithm corresponding to the range of eigenvalues is used as the recommendation algorithm corresponding to each user group.

Optionally, before determining the data feature value range to which the data feature value of each user group belongs in S401, the method may further include:

S401a. Determine the data characteristic value of each user group.

Optionally, the data characteristics of each user group include at least one of the following: historical record data of each user group, label data of each user group, attribute data of each user group, and each user group The social data of the group; the data characteristic value of each user group includes: the characteristic value corresponding to each data characteristic of each user group.

Specifically, if the data features of each user group include at least one of the following: historical record data of each user group, label data of each user group, and social data of each user group, then each The characteristic value corresponding to each data feature of the user group includes: the average value of the data feature of all users in each user group; if the data feature of each user group includes: the attribute data of each user group, Then the characteristic value corresponding to each data characteristic of each user group includes: the attribute data range of all users in each user group.

For example, if the data feature of each user group includes the historical record data of each user group, then the feature value of the historical record data of each user group is the historical record data of all users in each user group average value. The historical record data of each user group includes: at least one of all historical record data of each user group and all historical record data of each user group within a preset time period. All historical record data of each user group include: the sum of browsing history record data of all users of each user group, the sum of purchase history record data of all users of each user group, and the sum of the purchase history record data of all users of each user group At least one type such as the sum of the collection history data of all users.

If all the historical data of each user group include: the sum L of browsing history data of all users of each user group, the sum G of purchase history data of all users of each user group, and the sum G of all users of each user group If the number of users in each user group is N, the feature values of all historical record data of each user group can be obtained according to the following formula (1).

(L+G+K)/N formula (1)

The historical record data of each user group within the preset time period includes: the sum L' of browsing history record data of all users of each user group within the preset time period, all users of each user group in The sum G' of purchase history data in the preset time period, and the collection history data K' of all users in each user group in the preset time period, if each user in each user group The number is N, then the feature value of the historical record data of each user group within the preset time period can be obtained according to the following formula (2).

(L’+G’+K’)/N formula (2)

In this embodiment, the data characteristics of each user group include the historical record data of each user group as an example. The method for determining the feature value of the label data and the feature value of the social data of each user group is the same as above. The process of determining the characteristic value of the historical record data of each user group is similar and will not be repeated here.

For example, if the feature value V1 corresponding to all historical record data of each user group may be an average value of all historical record data of all users in each user group. The feature value V2 corresponding to the recent historical record data of each user group may be an average value of the recent historical record data of all users in each user group. The feature value V3 corresponding to the entry data of each user group may be an average value of all the entry data of each user group. The feature value V4 corresponding to the valid tag data of each user group may be an average value of the valid tag data of all users in each user group. The feature value V5 corresponding to the prominent tag data of each user group may be an average value of the prominent tag data of all users in each user group. The feature value V6 corresponding to the user attribute data of each user group may be represented by whether all users of each user group have user attribute data. The characteristic value V7 corresponding to the number of contacts of each user group may be an average value of the number of contacts of all users in each user group. The feature value V8 corresponding to the number of close contacts of each user group may be an average value of the number of close contacts of all users in each user group.

The determination of the characteristic value range to which the data characteristic value of each user group belongs in S401 as described above may be to compare the data characteristic value of each user group with the corresponding characteristic value threshold, and then determine the characteristic value range of each user group. The eigenvalue range to which the data eigenvalues belong. The 401 may be, for example, determining the characteristic value range to which the data characteristic value of each user group belongs according to Table 1 below.

Table 1

Optionally, as the recommendation algorithm corresponding to the feature value range in S402 described above as the recommendation algorithm corresponding to each user group may include:

According to the feature value range to which the data feature value of each user group belongs, the recommendation algorithm corresponding to the feature value range in the preset recommendation algorithm library is determined, and the determined recommendation algorithm is used as the recommendation algorithm corresponding to each user group.

Specifically, in this step, the feature value corresponding to the data feature value may be determined according to the feature value range to which the data feature value of each user group belongs, and the usage conditions of each algorithm in the preset recommendation algorithm library for the data feature value range is not applicable, and then other recommendation algorithms other than the inapplicable recommendation algorithm in the preset recommendation algorithm library are determined as the recommendation algorithm corresponding to the characteristic value range.

The preset recommendation algorithm library can include content-based recommendation algorithm, user-based collaborative filtering recommendation algorithm, item-based collaborative filtering recommendation, tag-based recommendation algorithm, matrix decomposition-based collaborative filtering recommendation algorithm, user attribute-based recommendation At least one recommendation algorithm among algorithms such as algorithm and social-based recommendation algorithm. The conditions for using the content-based recommendation algorithm may include: the data feature of each user of each user group has historical record data. The user-based collaborative filtering recommendation algorithm and the item-based collaborative filtering recommendation may include: the data feature of each user of each user group has historical record data, and the feature value of the historical record data of each user group is greater than Or equal to the preset characteristic value of historical record data. The conditions for using the tag-based recommendation algorithm include: the data feature of each user in each user group has tag data. The conditions for using the collaborative filtering recommendation algorithm based on matrix decomposition include: the data feature of each user of each user group has historical record data. The conditions for using the recommendation algorithm based on user attributes include: the data feature of each user of each user group has attribute data. The usage conditions of the social-based recommendation algorithm include: the data feature of each user of each user group has social data, and the feature value of the social data of each user group is greater than or equal to the preset social data feature value .

Optionally, in the above steps, according to the feature value range to which the data feature value of each user group belongs, determine the recommendation algorithm corresponding to the feature value range in the preset recommendation algorithm library, and use the determined recommendation algorithm as the feature value range for each user group. The recommendation algorithm corresponding to each user group may include:

The type of each user group is determined according to the feature value range to which the data feature value of each user group belongs; the recommendation algorithm corresponding to each user group is determined according to the recommendation algorithm corresponding to the type of each user group.

Wherein, determining the type of each user group may: determine the type of each user group according to the feature value range to which the data feature value of each user group belongs and Table 1 above; determine the recommendation algorithm corresponding to each user group It may be: according to the type of each user group and the usage conditions of each recommendation algorithm in the recommendation algorithm database for the type of user group, determine the recommendation algorithm corresponding to the type of each user group.

Wherein, the usage conditions of each recommendation algorithm in the recommendation algorithm library for the type of user group may be, for example, as shown in Table 2 below.

Table 2

"X" in Table 2 may be used to indicate that the current user group type is not suitable for the corresponding recommendation algorithm.

Optionally, if the recommendation algorithm corresponding to each user group includes multiple recommendation algorithms; the recommendation strategy includes: N kinds of fusion recommendation algorithms corresponding to each user group; wherein, each fusion recommendation algorithm includes the multiple recommendation algorithms A fusion method corresponding to the algorithm and the multiple recommendation algorithms.

Optionally, determining the recommendation strategy corresponding to each user group in S302 as described above may also include:

S403. According to the recommendation algorithm corresponding to each user group, and, the use conditions of each fusion method for the recommendation algorithm in the preset fusion method library, determine the recommendation algorithm corresponding to each user group from the recommendation algorithm library The N fusion ways of the usage conditions are the N fusion ways corresponding to the plurality of recommendation algorithms.

Specifically, the preset fusion method library may include: at least one fusion method among a weighted fusion method, a crossover fusion method, a hierarchical fusion method, and a waterfall fusion method.

Wherein, the weighted fusion method may be by weighting the scores of the recommendation information determined according to different recommendation algorithms, sorting the weighted recommendation information, determining the priority of the weighted recommendation information, and then according to The priority of the weighted recommendation information determines the corresponding recommendation information sent to each user group. Wherein, the weight used in the weighting can be dynamically adjusted according to the feedback information of each user in each user group on the recommendation information.

The cross fusion method may be to determine the recommendation information corresponding to each user group according to a preset ratio for the recommendation information determined according to different recommendation algorithms.

This hierarchical fusion method can determine the priority of the recommended information determined according to different recommendation algorithms according to the feedback information of each user in each user group for the recommended information, and then according to the priority of the recommended information, the highest priority The corresponding at least one recommendation information is determined as the recommendation information corresponding to each user group.

The waterfall fusion method may, for example, use the recommendation information determined by one recommendation algorithm as the input information of another recommendation algorithm, and then determine the recommendation information corresponding to each user group.

Wherein, the conditions for using the recommendation algorithm in the weighted fusion method may include: the recommendation information determined according to the recommendation algorithm needs to include prediction score information. If the recommendation information determined by a recommendation algorithm does not include predictive score information, the recommendation algorithm is not suitable for adopting the weighted fusion method with other recommendation algorithms.

The conditions for using the recommendation algorithm in the cross-type fusion method may include: the length of the recommendation information determined according to the recommendation algorithm must be greater than the length of the recommendation information corresponding to each user group to be generated.

The conditions for using the recommendation algorithm in the hierarchical fusion method may include: the length of the recommendation information determined according to the recommendation algorithm must be smaller than the length of the recommendation information corresponding to each user group to be generated.

The conditions for using the recommendation algorithm in the waterfall fusion method may include: the length of the recommendation information determined according to the recommendation algorithm must be greater than the length of the recommendation information corresponding to each user group to be generated.

S404. According to the N fusion methods, respectively fuse the multiple recommendation algorithms to obtain the N fusion recommendation algorithms.

Optionally, generating the recommendation information for each user group according to the recommendation strategy corresponding to each user group in S303 above may include:

S405. Determine the recommendation accuracy of each fusion recommendation algorithm corresponding to each user group.

S406. Determine the fusion recommendation algorithm with the highest recommendation accuracy from the N fusion recommendation algorithms as the optimal fusion recommendation algorithm for each user group.

S407. Generate recommendation information for each user group according to the optimal fusion recommendation algorithm for each user group.

Optionally, determining the recommendation accuracy of each fusion recommendation algorithm corresponding to each user group in S405 above may include:

For one of the N fusion recommendation algorithms corresponding to each user group, according to a set of data in the N sets of data of each user group, the recommendation information corresponding to the fusion recommendation algorithm is generated. Among the N sets of data Different groups of data correspond to the same number of users;

According to the user's feedback information on the recommendation information corresponding to the group of data, the recommendation accuracy of the fusion recommendation algorithm corresponding to each user group is determined.

Specifically, the N groups of data may be data randomly selected from all user data of each user group. The difference of the same data feature in the different groups of data is within a preset range. That is to say, the data characteristics of the different groups of data are similar, and the corresponding number of users is the same. Therefore, the N groups of data may be called N groups of equal data.

For example, FIG. 5 is a diagram of a correspondence relationship between a user group, a fusion recommendation strategy, a data group, and recommendation accuracy of recommendation information provided by an embodiment of the present invention. As shown in FIG. 5 , if the recommendation strategy corresponding to a user group includes: fusion recommendation algorithm 1 , fusion recommendation algorithm 2 , fusion recommendation algorithm 3 , fusion recommendation algorithm 4 and fusion recommendation algorithm 5 . The one user group includes 5 sets of data, that is, data set 1 , data set 2 , data set 3 , data set 4 and data set 5 . According to the data group 1 of this user group, the recommendation information corresponding to the fusion recommendation algorithm 1 can be obtained by adopting the fusion recommendation algorithm 1; the recommendation information corresponding to the fusion recommendation algorithm 2 can be obtained by using the fusion recommendation algorithm 2 according to the data group 2; The data group 3 adopts the fusion recommendation algorithm 3 to obtain the recommendation information corresponding to the fusion recommendation algorithm 3; according to the data group 4, adopts the fusion recommendation algorithm 4 to obtain the recommendation information corresponding to the fusion recommendation algorithm 4; according to the data group 5, adopts fusion The recommendation algorithm 5 can obtain the recommendation information corresponding to the fusion recommendation algorithm 5 . If the recommendation accuracy of the recommendation information corresponding to the fusion recommendation algorithm 1 is 0.8, the recommendation accuracy of the recommendation information corresponding to the fusion recommendation algorithm 2 is 0.3, and the recommendation accuracy of the recommendation information corresponding to the fusion recommendation algorithm 3 is 0.4, the The recommendation accuracy of the recommendation information corresponding to the fusion recommendation algorithm 4 is 0.9, and the recommendation accuracy of the recommendation information corresponding to the fusion recommendation algorithm 5 is 0.5. Therefore, the highest recommendation accuracy of this user group is 0.9, then the fusion recommendation algorithm 4. An algorithm may be recommended for the optimal fusion of the one user group.

Optionally, the method may also include:

If the proportion of the user group whose highest recommendation accuracy is less than the preset accuracy threshold is greater than or equal to the preset threshold, the multiple users are regrouped; the proportion of the user group whose highest recommendation accuracy is less than the preset accuracy threshold can be is the ratio of the number of user groups whose highest recommendation accuracy is less than the preset accuracy threshold to the number of the at least two user groups;

Determine the corresponding recommendation strategy for each user group after the regrouping;

generating recommendation information for each of the regrouped user groups according to the recommendation strategy corresponding to each of the regrouped user groups;

The recommendation information is sent to the terminals corresponding to the users in each of the regrouped user groups.

Optionally, determining the corresponding recommendation strategy for each user group after the regrouping may include:

Determine whether the data characteristic value of each user group after the regrouping is within the characteristic value range where the data characteristic value of the original grouped user group is located;

Determine whether the highest recommendation accuracy of the user group after the original grouping is greater than or equal to the preset accuracy threshold;

If the data characteristic value of each user group after the regrouping is within the characteristic value range of the data characteristic value of the original grouped user group, and the highest recommendation accuracy of the original grouped user group is greater than or equal to For the preset accuracy threshold, the recommendation strategy corresponding to the original grouped user group is used as the recommendation strategy corresponding to each user group after the regrouping.

The embodiment of the present invention can better ensure more accurate user grouping, more accurate determination of recommendation strategies, and better guarantee of information The recommendation algorithm obtains the accuracy of recommended information.

The information recommendation method provided by the embodiment of the present invention uses specific examples to illustrate the method described in the above embodiment, and its beneficial effect is similar to that of the above embodiment, so it will not be repeated here.

The embodiment of the present invention also provides a server. FIG. 6 is a schematic structural diagram of a server provided by an embodiment of the present invention. As shown in FIG. 6 , the server 600 may include: a grouping module 601 , a determining module 602 , a generating module 603 and a sending module 604 .

Wherein, the grouping module 601, the determining module 602, and the generating module 603 can be implemented by the processor 201 in the server as described in FIG. , the function of each module of the determination module 602 and the generation module 603 . The sending module 604 can be realized through the network interface 203 in the server as shown in FIG. 2 above.

Wherein, the grouping module 601 is configured to group multiple users to obtain at least two user groups; wherein, each user group includes at least one user.

The determining module 602 is configured to determine a recommendation strategy corresponding to each user group; the recommendation strategy includes a recommendation algorithm corresponding to each user group.

A generating module 603, configured to generate recommendation information for each user group according to a recommendation strategy corresponding to each user group.

A sending module 604, configured to send the recommendation information to a terminal corresponding to the at least one user.

Optionally, the grouping module 601 is configured to group the multiple users according to the data characteristics of each user in the multiple users to obtain the at least two user groups.

Optionally, the data feature of each user includes: at least one data feature.

The grouping module 601 is configured to group the multiple users according to the at least one data characteristic of each user according to a preset sequence of data characteristics to obtain the at least two user groups.

Optionally, the data characteristics of each user include at least one of the following:

The historical record data of each user, the tag data of each user, the attribute data of each user and the social data of each user.

Wherein, the historical record data of each user includes: at least one of all historical record data of each user and historical record data of each user within a preset time period; the label data of each user includes At least one of the number of valid tags for each user and the number of outstanding tags for each user. The attribute data of each user includes at least one of the gender of each user, the age group of each user, and the identity of each user. The social data of each user includes at least one of the number of contacts of each user, the number of close contacts of each user, and the number of active contacts of each user. The close contacts of each user are contacts whose contact times with each user are greater than or equal to a preset times threshold.

Optionally, the determining module 602 determines the characteristic value range to which the data characteristic value of each user group belongs, and uses the recommendation algorithm corresponding to the characteristic value range as the recommendation algorithm corresponding to each user group.

Optionally, the determining module 602 is also configured to determine the data characteristic value of each user group.

Optionally, the data characteristics of each user group include at least one of the following: historical record data of each user group, label data of each user group, attribute data of each user group, and each user group The social data of the group; the data characteristic value of each user group includes: the characteristic value corresponding to each data characteristic of each user group.

Optionally, the determination module 602 is further configured to determine the recommendation algorithm corresponding to the feature value range in the preset recommendation algorithm library according to the feature value range to which the data feature value of each user group belongs, and use the determined recommendation algorithm As the recommendation algorithm corresponding to each user group.

Optionally, if the recommendation algorithm corresponding to each user group includes multiple recommendation algorithms; the recommendation strategy includes: N kinds of fusion recommendation algorithms corresponding to each user group; wherein, each fusion recommendation algorithm includes the multiple recommendation algorithms A fusion method corresponding to the algorithm and the multiple recommendation algorithms.

The determining module 602 is further configured to determine from the recommendation algorithm library that each user group is compatible with the recommendation algorithm according to the recommendation algorithm corresponding to each user group, and the usage conditions of each fusion method in the preset fusion method library for the recommendation algorithm. The N fusion methods of the use conditions of the recommendation algorithm corresponding to the group are the N fusion methods corresponding to the multiple recommendation algorithms; according to the N fusion methods, the multiple recommendation algorithms are respectively fused to obtain the N fusion recommendation algorithms .

Optionally, the determination module 602 is also used to determine the recommendation accuracy of each fusion recommendation algorithm corresponding to each user group; determine the fusion recommendation algorithm with the highest recommendation accuracy from the N fusion recommendation algorithms as the fusion recommendation algorithm for each user group. An optimal fusion recommendation algorithm for a user group;

The generation module 603 is further configured to generate recommendation information for each user group according to the optimal fusion recommendation algorithm for each user group.

The server provided by the embodiment of the present invention can be used to implement the information recommendation method described in the above embodiment, and the beneficial effect is similar to that of the above embodiment, so it will not be repeated here.

The embodiment of the present invention also provides a computer-readable storage medium. FIG. 7 is a schematic structural diagram of a computer-readable storage medium provided by an embodiment of the present invention.

As shown in FIG. 7 , the computer-readable storage medium 700 can store executable program code 701 . The program code 701 is used to implement the information recommendation method described in any one of the first and second embodiments in the above claims 1-10.

The computer-readable storage medium provided by the embodiments of the present invention can store program codes that can be used to implement the information recommendation method described in the above-mentioned embodiments, and the beneficial effects are similar to those of the above-mentioned embodiments, so details are not repeated here.

Those of ordinary skill in the art can understand that all or part of the steps for implementing the above method embodiments can be completed by program instructions and related hardware. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above-mentioned method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other various media that can store program codes.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: It is still possible to modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the various embodiments of the present invention. scope.

Claims (21)

1. An information recommendation method, characterized in that, comprising: Grouping multiple users to obtain at least two user groups; wherein, each user group includes at least one user; Determining a recommendation strategy corresponding to each user group; the recommendation strategy includes: a recommendation algorithm corresponding to each user group; generating recommendation information for each user group according to a recommendation strategy corresponding to each user group; Sending the recommendation information to a terminal corresponding to the at least one user. 2. The method according to claim 1, wherein said grouping a plurality of users to obtain at least two user groups comprises: The multiple users are grouped according to the data characteristics of each of the multiple users to obtain the at least two user groups. 3. The method according to claim 2, wherein the data characteristics of each user comprise: at least one data characteristic; The step of grouping the multiple users according to the data characteristics of each of the multiple users to obtain the at least two user groups includes: According to the at least one data feature of each user, the multiple users are grouped according to a preset sequence of data features to obtain the at least two user groups. 4. The method according to claim 2 or 3, wherein the data characteristics of each user include at least one of the following: The historical record data of each user, the tag data of each user, the attribute data of each user and the social data of each user; Wherein, the historical record data of each user includes: at least one of all historical record data of each user and historical record data of each user within a preset time period; The tag data of each user includes: at least one of the number of valid tags for each user and the number of prominent tags for each user; The attribute data of each user includes at least one of the gender of each user, the age group of each user, and the identity of each user; The social data of each user includes at least one of the number of contacts of each user, the number of close contacts of each user, and the number of active contacts of each user; The close contact is a contact whose number of contacts with each user is greater than or equal to a preset number of times threshold. 5. The method according to any one of claims 1-4, wherein the determining the recommendation strategy corresponding to each user group comprises: determining the characteristic value range to which the data characteristic value of each user group belongs; The recommendation algorithm corresponding to the feature value range is used as the recommendation algorithm corresponding to each user group. 6. The method according to claim 5, wherein before the determination of the characteristic value range to which the data characteristic value of each user group belongs, the method further comprises: Determine the data characteristic value of each user group. 7. The method according to claim 6, wherein the data characteristics of each user group include at least one of the following: historical record data of each user group, label data of each user group , the attribute data of each user group and the social data of each user group; the data characteristic value of each user group includes: a characteristic value corresponding to each data characteristic of each user group. 8. The method according to claim 6 or 7, wherein said using the recommendation algorithm corresponding to the feature value range as the recommendation algorithm corresponding to each user group comprises: According to the feature value range to which the data feature value of each user group belongs, determine the recommendation algorithm corresponding to the feature value range in the preset recommendation algorithm library, and use the determined recommendation algorithm as the corresponding to each user group recommendation algorithm. 9. The method according to any one of claims 1-8, wherein if the recommendation algorithm corresponding to each user group includes multiple recommendation algorithms; the recommendation strategy includes: each user group N kinds of corresponding fusion recommendation algorithms; wherein, each fusion recommendation algorithm includes the plurality of recommendation algorithms and a fusion method corresponding to the plurality of recommendation algorithms; The determining the recommendation strategy corresponding to each user group includes: According to the recommendation algorithm corresponding to each user group, and, the use conditions of each fusion method in the preset fusion method library for the recommendation algorithm, determine from the fusion method library that meets the recommendation corresponding to each user group The N fusion modes of the usage conditions of the algorithm are the N fusion modes corresponding to the plurality of recommendation algorithms; According to the N fusion methods, the plurality of recommendation algorithms are respectively fused to obtain the N fusion recommendation algorithms. 10. The method according to claim 9, wherein the generating the recommendation information of each user group according to the recommendation strategy corresponding to each user group comprises: Determining the recommendation accuracy of each fusion recommendation algorithm corresponding to each user group; Determine the fusion recommendation algorithm with the highest recommendation accuracy from the N fusion recommendation algorithms as the optimal fusion recommendation algorithm for each user group; Generate recommendation information for each user group according to the optimal fusion recommendation algorithm for each user group. 11. A server, characterized in that, comprising: A grouping module, configured to group multiple users to obtain at least two user groups; wherein, each user group includes at least one user; A determining module, configured to determine a recommendation strategy corresponding to each user group; the recommendation strategy includes: a recommendation algorithm corresponding to each user group; A generation module, configured to generate recommendation information for each user group according to a recommendation strategy corresponding to each user group; A sending module, configured to send the recommendation information to a terminal corresponding to the at least one user. 12. The server according to claim 11, wherein: The grouping module is configured to group the multiple users according to the data characteristics of each of the multiple users to obtain the at least two user groups. 13. The server according to claim 12, wherein the data characteristics of each user comprise: at least one data characteristic; The grouping module is configured to group the multiple users according to the at least one data characteristic of each user according to a preset sequence of data characteristics to obtain the at least two user groups. 14. The server according to claim 12 or 13, wherein the data characteristics of each user include at least one of the following: The historical record data of each user, the tag data of each user, the attribute data of each user and the social data of each user; Wherein, the historical record data of each user includes: at least one of all historical record data of each user and historical record data of each user within a preset time period; The tag data of each user includes: at least one of the number of valid tags for each user and the number of prominent tags for each user; The attribute data of each user includes at least one of the gender of each user, the age group of each user, and the identity of each user; The social data of each user includes at least one of the number of contacts of each user, the number of close contacts of each user, and the number of active contacts of each user; The close contact is a contact whose number of contacts with each user is greater than or equal to a preset number of times threshold. 15. The server according to any one of claims 11-14, characterized in that, The determining module is configured to determine the feature value range to which the data feature value of each user group belongs, and use the recommendation algorithm corresponding to the feature value range as the recommendation algorithm corresponding to each user group. 16. The server according to claim 15, wherein: The determining module is further configured to determine the data characteristic value of each user group. 17. The server according to claim 16, wherein the data characteristics of each user group include at least one of the following: historical record data of each user group, label data of each user group , the attribute data of each user group and the social data of each user group; the data characteristic value of each user group includes: a characteristic value corresponding to each data characteristic of each user group. 18. The server according to claim 16 or 17, wherein: The determination module is further configured to determine the recommendation algorithm corresponding to the range of feature values in the preset recommendation algorithm library according to the feature value range to which the feature value of the data of each user group belongs, and use the determined recommendation algorithm As the recommendation algorithm corresponding to each user group. 19. The server according to any one of claims 11-18, wherein if the recommendation algorithm corresponding to each user group includes multiple recommendation algorithms; the recommendation strategy includes: each user group N kinds of corresponding fusion recommendation algorithms; wherein, each fusion recommendation algorithm includes the plurality of recommendation algorithms and a fusion method corresponding to the plurality of recommendation algorithms; The determination module is further configured to determine from the recommendation algorithm library that meets the recommended algorithm according to the recommendation algorithm corresponding to each user group, and the usage conditions of each fusion method in the preset fusion method library for the recommendation algorithm. The N fusion methods of the use conditions of the recommendation algorithm corresponding to each user group are the N fusion methods corresponding to the multiple recommendation algorithms; according to the N fusion methods, the multiple recommendation algorithms are respectively fused, The N fusion recommendation algorithms are obtained. 20. The server according to claim 19, wherein: The determination module is also used to determine the recommendation accuracy of each fusion recommendation algorithm corresponding to each user group; determine the fusion recommendation algorithm with the highest recommendation accuracy from the N fusion recommendation algorithms as the An optimal fusion recommendation algorithm for a user group; The generation module is further configured to generate recommendation information for each user group according to the optimal fusion recommendation algorithm for each user group. 21. A server, characterized by comprising a processor, a memory, a communication interface, and a bus; wherein, the processor is connected to the memory and the communication interface through the bus; The memory is used to store instructions; The processor is configured to execute instructions, and when the processor executes the instructions stored in the memory, the processor is made to execute the information recommendation method described in any one of claims 1-10 above.