TWI818213B - System and method of product recommendation and computer readable medium - Google Patents

Abstract

A product recommendation method and its corresponding system are provided. The product recommendation method includes: providing a dialog interface, interacting with the user through the dialog interface and receiving the user’s dialog text, extracting at least one keyword in the dialog text, calculating a recommendation priority of each keyword, and providing recommended products related to each keyword to the user according to the recommendation priority. The product recommendation method also includes: collecting each user’s at least one online product-related purchase records, advertisement clicking records and browsing records, collecting each user’s behavior records and product interaction records in a physical store, selecting a target user among these users, and providing recommended products to the target user based on the purchase records, the advertisement clicking records, the browsing records, the behavior records, and the product interaction records. The present invention further provides a computer-readable medium for performing the product recommendation method.

Description

Product recommendation system and method and computer-readable medium

The present invention relates to a product recommendation technology, and in particular to a product recommendation system and method based on keywords or user behavior.

In the past, companies used passive marketing. Although there were many services and products that could provide users with more convenience, they had to wait until users arrived at the store in need before store staff could introduce or recommend products to users. This resulted in low sales rates and low corporate revenue. If the scale increases, it will also fall behind other players when occupying the market, which will have an adverse impact on the company's human, financial, marketing and market deployment.

Therefore, there is a need for a technology that can effectively, quickly, accurately and proactively provide users with more convenient services and products to improve the effectiveness of product recommendations.

In order to solve the above problems, the present invention provides a product recommendation system, including: an artificial intelligence interactive module for providing a dialogue interface, and conducting dialogue and interaction with the user through the dialogue interface to receive the user's dialogue text; and keyword recommendation. Module used to capture the text of the conversation If one keyword is missing, the recommendation order of each keyword is calculated to provide the user with recommended products of each keyword based on the recommendation order.

The present invention also provides a product recommendation system, including: a plurality of record modules, used to collect purchase records, advertisement click records and browsing records related to at least one product on the Internet for each user of the product recommendation system; a store data module, Used to collect the behavior records and product interaction records of each user in the physical store of the product recommendation system; and the hybrid recommendation module, used to select target users among the users of the product recommendation system to base on the purchase records , the advertisement click record, the browsing record, the behavior record and the product interaction record, and provide recommended products to the target user.

The present invention also provides a product recommendation method, which includes: providing a dialogue interface; conducting dialogue and interaction with the user through the dialogue interface to receive the user's dialogue text; retrieving at least one keyword in the dialogue text to calculate each key the recommended order of words; and provide recommended products for each keyword to the user based on the recommended order.

The present invention further provides a product recommendation method, which includes: collecting the purchase records, advertisement click records and browsing records of each user on the Internet related to at least one product; collecting the behavior records and product interaction records of each user in the physical store; A target user is selected among these users to provide recommended products to the target user based on the purchase record, the ad click record, the browsing record, the behavior record and the product interaction record.

The present invention further provides a computer-readable medium, which is used in a computing device or a computer and stores instructions to execute the above product recommendation method.

The product recommendation system and method and the computer-readable medium of the present invention improve the overall sales method, so that store staff no longer passively recommend products to users, but can instead actively recommend products. In addition, by clearly understanding the user information content and user recommendation content, the overall sales performance can be greatly improved. Achievements. Furthermore, if the user cannot go to the physical store, product recommendation information will be pushed to the user through the system, and the original passive marketing will be changed to active marketing, which can quickly discover target users, quickly occupy market segments, and improve sales strategies.

100: Product recommendation system

101:User

102:Mobile device

103: Electronic devices

104:Physical store

110: Artificial intelligence interactive module

121: Purchase record module

122: Advertising click module

123: Browsing history module

124: Positioning module

125: Biometric Module

126: Store data module

130,140:Database

150:Hybrid recommendation module

151:Keyword recommendation module

152: User similarity recommendation module

153: Product similarity recommendation module

401~404: User

451~454: Commodity

601:User

651~653: Commodity

S210~S270, S310~S360, S510~S560: Method steps

Figure 1 is a schematic block diagram of a product recommendation system according to an embodiment of the present invention.

Figure 2 is a keyword recommendation flow chart of a product recommendation method according to an embodiment of the present invention.

Figure 3 is a user similarity recommendation flow chart of a product recommendation method according to an embodiment of the present invention.

Figure 4 is an example diagram of user similarity recommendation according to a product recommendation method according to an embodiment of the present invention.

Figure 5 is a product similarity recommendation flow chart of a product recommendation method according to an embodiment of the present invention.

Figure 6 is an example diagram of product similarity recommendation according to a product recommendation method according to an embodiment of the present invention.

The following describes the implementation of the present invention through specific embodiments. Those with ordinary knowledge in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification.

FIG. 1 is a schematic block diagram of a product recommendation system 100 according to an embodiment of the present invention. The product recommendation system 100 includes an artificial intelligence interaction module 110, a purchase record module 121, an advertisement click module 122, a browsing record module 123, a positioning module 124, a biometric module 125, a store data module 126, and a database 130 and 140, and a hybrid recommendation module 150, wherein the hybrid recommendation module 150 includes a keyword recommendation module 151, a user similarity recommendation module 152, and a product similarity recommendation module 153. The database 130 communicates with the artificial intelligence interactive module 110 and the keyword recommendation module 151. The database 140 communicates with the purchase record module 121, advertisement click module 122, browsing record module 123, positioning module 124, biometric module 125, store information module 126, user similarity recommendation module 152, and products Similarity recommendation module 153.

Each module of the product recommendation system 100 in Figure 1 can be software, hardware or firmware; if it is hardware, it can be a processing unit, processor, computer or server with data processing and computing capabilities; if If it is software or firmware, it may include instructions executable by a processing unit, processor, computer or server. Each module and each database of the product recommendation system 100 in Figure 1 can be integrated in the same hardware device, or can be distributed and built in multiple hardware devices.

The product recommendation system 100 can be affiliated to a telecommunications company or merchant that operates both online stores and physical stores, and is used to recommend products to users. The above-mentioned goods are physical products that can be purchased (such as various electronic products or consumer products) or services that can be applied for (such as telecommunications services, medical services or insurance services). The artificial intelligence interaction module 110, the purchase record module 121, the advertisement click module 122, the browsing record module 123, the positioning module 124, the biometric module 125, and the store information module 126 are used to collect information from the user's mobile device and Computers and other electronic devices (such as the mobile device 102 and electronic device 103 of the user 101) collect various data and store the data in the databases 130 and 140 respectively. The keyword recommendation model The group 151, the user similarity recommendation module 152, and the product similarity recommendation module 153 are used to execute the product recommendation methods shown in Figure 2, Figure 3, and Figure 5 respectively based on the data.

Figure 2 is a keyword recommendation flow chart of a product recommendation method according to an embodiment of the present invention.

First, in step S210, the user 101 can use the mobile device 102 or the electronic device 103 to open the dialogue interface provided by the artificial intelligence interactive module 110, and use text input to engage in dialogue and interaction with the artificial intelligence interactive module 110 to learn about the telecommunications company or the Various products provided by merchants. The artificial intelligence interaction module 110 uses the trained artificial intelligence model and interacts with the user 101 through the conversation interface.

In step S220 , the artificial intelligence interaction module 110 receives text input by the user 101 during the dialogue interaction process (hereinafter referred to as dialogue text) through the dialogue interface, so as to store the dialogue text in the database 130 .

Next, in step S230, the keyword recommendation module 151 obtains the dialogue text of the user 101 from the database 130, and then filters out special symbols such as punctuation marks and emoticons in the dialogue text, and filters out particles and exclamations in the dialogue text. words and other interference words.

In step S240, the keyword recommendation module 151 retrieves at least one keyword in the filtered dialogue text, and then calculates the occurrence _frequency tfi of each keyword i using the following formula.

The user's dialogue text may include plural sentences, in which the text sent by the user each time he clicks the send button of the dialogue interface during the dialogue interaction process is one sentence. In the above formula, n _i is the number of sentences containing the keyword i in the dialogue text of user 101, and N is the number of all sentences in the dialogue text of user 101. In other words, the occurrence frequency tfi of each keyword i is calculated and generated based on the occurrence ratio of each keyword i in the dialogue text _of the user 101.

Next, in step S250, the keyword recommendation module 151 calculates _the importance idfi of each keyword i in the filtered dialogue text using the following formula.

The sentences in the user's dialogue text can be divided into plural segments, and each segment includes a plurality of sentences. The number of sentences included in each segment may be the same or different. Each segment can be stored in any form such as a file. In the above formula, D is the section of all dialogue texts of all users stored in the database 130 (not limited to user 101, but all users of telecommunications companies or merchants, these users can also be regarded as all users of the product recommendation system 100) The total number of segments, and d _i is the total number of segments containing the keyword i among all dialogue texts of all users stored in the database 130 . In other words, the importance idf i of each keyword _i is calculated based on the proportion of occurrence of each keyword i in the dialogue text of all users.

Next, in step S260, the keyword recommendation module 151 calculates the recommendation order Ri of each keyword i in the filtered dialogue text using the following _formula .

R _i = tf _i × idf _i

Next, in step S270, the keyword recommendation module 151 provides and recommends at least one product related to the keyword i to the user 101 according to the recommendation order R _i of each keyword i . Specifically, in the plural keywords of the dialogue text, when the value of the recommendation order R _i of a certain keyword i is higher, the recommendation order of the related products of the keyword i is more priority. The keyword recommendation module 151 can provide the recommendation information of the above-mentioned related products to the store personnel for reference, and can also transmit the above-mentioned recommendation information of the related products to the mobile device 102 and/or the electronic device 103 of the user 101 .

For example, user 101 inputs "I want to know about the latest mobile phones" during the dialogue and interaction with the artificial intelligence interaction module 110, and mentions the keyword "mobile phone" multiple times during the subsequent dialogue and interaction, then through the above formula After calculation, the keyword recommendation module 151 can determine that the user 101 may be interested in the mobile phone product, and send the recommendation information of the mobile phone product to the mobile device 102 and/or the electronic device 103 of the user 101.

Figure 3 is a user similarity recommendation flow chart of a product recommendation method according to an embodiment of the present invention.

First, in step S310, the purchase record module 121, the advertisement click module 122, the browsing record module 123, and the store data module 126 respectively collect the purchase records, advertisement click records, and browsing records of each user of the telecommunications company or merchant. and store information, and store these records and information in the database 140. Specifically, the purchase record of each user is the record of the user purchasing or applying for products of the telecommunications company or merchant on the Internet. Each user's advertisement click record is a record of the user clicking on the product advertisements of the telecommunications company or merchant on the website or mobile device application. Each user's browsing history includes records of web pages that the user has searched for and/or browsed on the Internet related to the products of the telecommunications company or merchant.

Each user's store information includes the user's customer attributes, as well as the user's behavior records and product interaction records in the physical store (such as the physical store 104) of the telecommunications company or merchant. The customer attributes may include basic information such as gender, age, income, and interests of the user, and may be filled in by the user at the online or physical store of the telecommunications company or merchant. The behavior record may include the user's inquiry of product information, purchase of products and/or application for products in the physical store of the telecommunications company or merchant. record. In a physical store, this behavior can be recorded by a service person or automated service equipment. The product interaction record may include records of the user's interaction with the products displayed in the physical store, for example, which products the user has tried, or which products the user has viewed and the viewing duration exceeded the preset value. The above product interaction indicates that the user May be interested in items to try or view.

In one embodiment, cameras and audio recorders are installed in the physical store to record the user's images and sounds in the physical store. The store data module 126 can obtain the user's images and sounds in the physical store from the cameras and audio recorders of the physical store, and store the images and sounds in the database 140 . The store data module 126 can analyze the images and sounds to capture the user's product interaction records in the physical store, and then store the product interaction records in the database 140 . In addition, the store data module 126 can also analyze the images and sounds to capture the user's behavior records in the physical store, and then store the behavior records in the database 140 as records for service personnel or automatic service equipment. add.

In order to identify each user in these images and sounds, the service personnel or automatic service equipment of the physical store can record the time and location (for example, which physical store) each user arrives at the physical store, and the biometric module 125 can be based on the Obtain the clips of each user appearing in the images and sounds at the time and place, and extract the biometric characteristics such as voiceprint, handwriting, face shape and gait of each user from these clips, and then store the biometric characteristics of each user into the data. Library 140. The store data module 126 can also obtain the clips in which each user appears in the images and sounds based on the time and location, and then identify each user in the clips based on the biometric characteristics, and analyze the clips to retrieve Product interaction records and behavior records of each user.

In addition, the positioning module 124 is used to obtain the user's location information. If the product recommendation system 100 belongs to a telecommunications company, the positioning module 124 can perform a triangulation positioning operation based on the signal strength of the user's mobile device received by the telecommunications company's base station to obtain the user's location information, and This location information is stored in the database 140. Alternatively, regardless of whether the product recommendation system 100 is affiliated with a telecommunications company or a merchant, the positioning module 124 can also obtain the user's location information from the user's mobile device through the user's authorization, and store the location information in the database 140 . The location information can be used as a supplement to the time and location of each user arriving at the physical store recorded by the service personnel or automatic service equipment of the physical store. In other words, the biometric module 125 and the store data module 126 can obtain each user based on the location information. The fragments of users appearing in these images and sounds are then used to capture each user's biometric characteristics, product interaction records and behavior records.

Next, in step S320, the user similarity recommendation module 152 selects the target user u among all users of the telecommunications company or merchant, and selects the product i to be recommended. For example, the target user u can be a user who has recently purchased or applied for a product. The user similarity recommendation module 152 can learn which users have recently purchased or applied for a product from the purchase records provided by the purchase record module 121 . Alternatively, the target user u may be the user targeted by the promotional activity. The product i to be recommended may be a product targeted by promotional activities.

Next, in steps S330 to S360, the user similarity recommendation module 152 obtains the purchase records and advertisements provided by the purchase record module 121, the advertisement click module 122, the browsing record module 123 and the store information module 126 from the database 140. Click records, browsing records and store data, enter these records and data into the collaborative filtering (Collaborative Filtering) model for filtering, and use the user similarity matrix (User-User Similarity Matrix) to find the user set similar to the target user u , and then The products most frequently purchased by similar users are recommended to the target user u . The algorithm is detailed as follows.

In step S330, the user similarity recommendation module 152 generates a set S ( u ) of users with similar interests to the target user u . In one embodiment, the user similarity recommendation module 152 can use the following Jaccard similarity formula (Jaccard Index) to determine whether the interests of two users α and β are similar.

In the above Jacquard similarity formula, A and B are the collection of products for which users α and β have had interest behavior respectively. Among them, interest behavior means that the user has purchased or applied for a product, clicked on the advertisement, search or search for the product. Browsed web pages related to the product, or had behavior records or product interaction records related to the product in a physical store. The user similarity recommendation module 152 can learn any information based on the purchase records, ad click records, browsing records and store information provided by the purchase record module 121, ad click module 122, browsing record module 123 and store data module 126. Whether a user has the above-mentioned interest behavior for any product, and then generates a collection of products for which any user has interest behavior. If the interest similarity J ( α,β ) calculated by the Jacquard similarity formula is greater than the preset value (for example, 0.5), it is determined that the interests of users α and β are similar. The user similarity recommendation module 152 can substitute the set of products for which the target user u and each other user has interested behavior into the Jacquard similarity formula to calculate the interest similarity between the two users, and will make the interest similarity greater than the By default, other users join the set S ( u ).

In step S340, the user similarity recommendation module 152 generates a set N ( i ) of users who have had the above-mentioned interest behavior for the recommended product i .

In step S350, the user similarity recommendation module 152 calculates the recommendation score ρ ( u,i ) of the product i to be recommended for the target user u using the following user similarity matrix formula.

In the above formula, v is each user in the intersection of sets S ( u ) and N ( i ), w _uv is the interest similarity between target user u and user v . This interest similarity can be calculated using the above Jacquard similarity formula. Calculated, r _vi is user v ’s interest in product i to be recommended. The degree of interest r _vi is related to the interest behavior of user v for the product i to be recommended. Specifically, a corresponding weight value can be set for each of the above interest behaviors, and then each user v has had for the product i to be recommended. The interest degree r _vi can be obtained by adding the weight values corresponding to an interesting behavior.

In step S360, the user similarity recommendation module 152 determines whether to recommend the target user u according to the recommendation score ρ ( u,i ) and provides the product i to be recommended. Specifically, if the recommendation score ρ ( u,i ) is greater than a preset value (for example, 0.25), the user similarity recommendation module 152 recommends the product i to be recommended to the target user u . The user similarity recommendation module 152 can provide the recommendation information of the product i to be recommended to the store personnel for reference, and can also transmit the recommendation information of the product i to be recommended to the mobile device and/or electronic device of the target user u . On the contrary, if the recommendation score ρ ( u,i ) is less than or equal to the preset value, the user similarity recommendation module 152 does not recommend the product i to be recommended to the target user u , and the user similarity recommendation module 152 can select another product to be recommended. Repeat the above steps for the product.

Figure 4 is an example diagram of user similarity recommendation according to a product recommendation method according to an embodiment of the present invention, in which the solid line represents "purchase" and the dotted line represents "recommendation". As shown in Figure 4, user 401 purchased products 451, 452 and 454, user 402 purchased product 454, user 403 purchased product 453, and target user 404 purchased products 451 and 452. After the above process, the user similarity recommendation module 152 determines that the interests of the user 401 and the target user 404 are similar, and selects the product 454 that the user 401 has purchased and the target user 404 has not purchased as the product to be recommended. Then, because the product 454 to be recommended is The recommendation score of the target user 404 is greater than the preset value, so the user similarity recommendation module 152 recommends the product 454 to the target user 404 .

Figure 5 is a product similarity recommendation flow chart of a product recommendation method according to an embodiment of the present invention.

First, in step S510, the purchase record module 121, the advertisement click module 122, the browsing record module 123, and the store data module 126 collect data from each user of the telecommunications company or merchant respectively. Purchase records, advertisement click records, browsing records and store information, and store these records and information in the database 140. Step S510 is the same as step S310 in Figure 3, and details will not be described again.

Next, in step S520, the product similarity recommendation module 153 selects the target user u among all users of the telecommunications company or merchant.

Next, in steps S530 to S560, the product similarity recommendation module 153 obtains the purchase records and advertisements provided by the purchase record module 121, the advertisement click module 122, the browsing record module 123 and the store data module 126 from the database 140. Click records, browsing records and store data, input these records and data into the collaborative filtering (Collaborative Filtering) model for filtering, and use the item similarity matrix (Item-Item Similarity Matrix) to find products similar to those recently browsed or purchased by target user u After collecting the products, similar products are recommended to the target user u . The algorithm is detailed as follows.

In step S530, the product similarity recommendation module 153 generates a set M ( u ) of products in which the target user u has interested behavior. As mentioned above, the product similarity recommendation module 153 can be based on the purchase records, advertisement click records, browsing records and store information provided by the purchase record module 121, the advertisement click module 122, the browsing record module 123 and the store information module 126. The data knows whether any user has any interest behavior in any product, and then generates a collection of products in which any user has interest behavior.

In step S540, the product similarity recommendation module 153 generates a set S ( j ) of products similar to the product j purchased by the target user u . Here, the similarity of two products means that the two products have the same label. Specifically, each product can have multiple labels to indicate the product's brand, place of origin, whether it is an organic product, etc., if among all labels on two products, the number of identical labels is divided by the number of all labels. If the value is greater than the preset value (for example, 0.5), the product similarity recommendation module 153 determines that the two products are similar. In this way, the product similarity recommendation module 153 can generate products similar to the product j that the target user u has purchased. The set S ( j ).

In another embodiment, the above-mentioned product j can also be a product that the target user u has had an interest in (such as browsing or purchasing, etc.), so the set S ( j ) can also be a product that the target user u has had an interest in. A collection of products similar to product j .

In step S550, the product similarity recommendation module 153 calculates the recommendation score P ( u,i ) of the product to be recommended using the following product similarity matrix formula.

In the above formula, product i is each product in the intersection of set M ( u ) and set S ( j ), w _ij is the similarity between product i and product j , r _ui is the target user u's relationship with product i degree of interest. The product similarity recommendation module 153 can calculate the similarity w _ij between product i and product j using the following formula. The calculation of the interest degree r _ui can be compared with the aforementioned interest degree r _vi .

In the above formula, N ( i ) and N ( j ) are the sets of users who have shown interest in product i and product j respectively.

In step S560, the product similarity recommendation module 153 determines whether to provide the recommended product i to the target user u based on the recommendation score P ( u,i ). Specifically, if the recommendation score P ( u,i ) is greater than the preset value (for example, 0.25), the product similarity recommendation module 153 recommends the intersection of the set M ( u ) and the set S ( j ) to the target user u Each item i . The product similarity recommendation module 153 can provide the recommendation information of product i to store personnel for reference, and can also transmit the recommendation information of product i to the mobile device and/or electronic device of target user u . On the contrary, if the recommendation score P ( u,i ) is less than or equal to the preset value, the product similarity recommendation module 153 does not recommend product i to target user u .

Figure 6 is an example diagram of product similarity recommendation according to a product recommendation method according to an embodiment of the present invention, in which the solid line represents "purchase" and the dotted line represents "recommendation". As shown in Figure 6, target user 601 purchased products 651 and 652. After the above calculation, the product similarity recommendation module 153 recommends the product 653 that is similar to the products 651 and 652 to the target user 601. For example, products 651 and 652 are computers and keyboards respectively, and product 653 is a mouse.

In addition, the present invention also discloses a computer-readable medium, which is applied to a computing device or computer having a processor (eg, CPU, GPU, etc.) and/or a memory, and stores instructions, and can utilize the computing device or computer. The computer executes the computer-readable medium through the processor and/or memory to perform the above methods and steps when executing the computer-readable medium.

To sum up, the product recommendation system and method and the computer-readable medium proposed by the present invention have the following characteristics and effects:

1. The artificial intelligence interactive module can analyze the user's semantic intention and analyze the products that the user may be interested in and the service information that may be needed through dialogue keywords. The product recommendation system and method of the present invention can recommend products based on this. This method improves the success rate of selling goods.

2. If the customer has purchased or applied for a product, the product recommendation system and method of the present invention can use the user's purchase records, advertising click records, browsing records, and store data analysis to calculate the similarity between users and products, and recommend similar products to similar users. Purchase products or recommend other products that are most similar to the current product, and perform cross-selling. After converting into user groups with similar interests, view their most frequently purchased products and recommend them to the currently locked target users. .

3. The product recommendation system and method of the present invention can obtain the user's location information, thereby analyzing the user's past locations and the stores where the user may purchase, and combining the user's browsing history and biometric data to find the products most similar to the currently browsed products. Product groups are then recommended to users.

4. Store staff can obtain real-time user needs from the product recommendation system and method of the present invention, so as to avoid recommending unsuitable products to users, resulting in poor user experience and affecting service quality. On the contrary, the product sales rate can be increased in this way , and increase overall revenue.

The above embodiments are only illustrative to illustrate the principles and effects of the present invention, but are not intended to limit the present invention. Anyone with ordinary knowledge in this technical field can modify and change the above embodiments without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention should be as listed in the patent application scope described below.

100: Product recommendation system

101:User

102:Mobile device

103: Electronic devices

104:Physical store

110: Artificial intelligence interactive module

121: Purchase record module

122: Advertising click module

123: Browsing history module

124: Positioning module

125: Biometric Module

126: Store data module

130,140:Database

150:Hybrid recommendation module

151:Keyword recommendation module

152: User similarity recommendation module

153: Product similarity recommendation module

Claims (9)

A product recommendation system includes: an artificial intelligence interaction module for providing a dialogue interface, and conducting dialogue interaction with a user through the dialogue interface to receive the user's dialogue text; and a keyword recommendation module for retrieving the user's dialogue text. For at least one keyword in the dialogue text, calculate the frequency of occurrence of each keyword based on the proportion of occurrence of each keyword in the user's dialogue text, and calculate the frequency of occurrence of each keyword based on the dialogue of all users of the product recommendation system based on each keyword The importance of each keyword is calculated based on the occurrence ratio in the text, and then the recommendation order of each keyword is calculated based on the frequency of occurrence and the importance, so as to provide the user with recommended products for each keyword based on the recommendation order. The product recommendation system as described in request item 1 includes: a plurality of record modules, used to collect purchase records, advertisement click records and browsing records related to at least one product on the Internet for each user of the product recommendation system; a store data module The group is used to collect the behavior records and product interaction records of each user in the physical store of the product recommendation system; and the user similarity recommendation module is used to select target users among the users of the product recommendation system to Based on the purchase record, the ad click record, the browsing record, the behavior record and the product interaction record, provide the target user with the recommended products that are of interest to users with similar interests as the target user. The product recommendation system as described in request item 1 further includes: a plurality of record modules, used to collect purchase records, advertisement click records and browsing records related to at least one product on the Internet for each user of the product recommendation system; The store data module is used to collect the behavior records and product interaction records of each user in the physical store of the product recommendation system; and the product similarity recommendation module is used to select targets among the users of the product recommendation system. The user can provide the target user with recommended products that are similar to the products that the target user has been interested in based on the purchase record, the ad click record, the browsing record, the behavior record and the product interaction record. A product recommendation system, including: a plurality of record modules, used to collect purchase records, advertisement click records and browsing records related to at least one product on the Internet for each user of the product recommendation system; a store data module, used to collect the product The recommendation system records each user's behavior in the physical store, obtains the user's images and sounds in the physical store from the camera and recorder of the physical store, and then analyzes the images and sounds to capture the user's behavior in the physical store. The product interaction record in the physical store, the product interaction record includes which products in the physical store the user has viewed and the viewing duration exceeds the preset value; and a hybrid recommendation module used in the product recommendation system Select a target user among other users to provide recommended products to the target user based on the purchase record, the ad click record, the browsing record, the behavior record and the product interaction record. The product recommendation system as described in request item 4, wherein the hybrid recommendation module includes: a user similarity recommendation module, used to base on the purchase record, the ad click record, the browsing record, the behavior record and the product interaction Record and provide the target user with the recommended products that have been of interest to users with similar interests to the target user; and The product similarity recommendation module is used to provide the target user with products similar to the products that the target user has been interested in based on the purchase record, the ad click record, the browsing record, the behavior record and the product interaction record. This recommended product. A product recommendation method, including: providing a dialogue interface; conducting dialogue and interaction with a user through the dialogue interface to receive the user's dialogue text; retrieving at least one keyword in the dialogue text; and selecting a keyword among the user based on each of the keywords. Calculate the frequency of occurrence of each keyword based on the proportion of occurrence in the dialogue text; calculate the importance of each keyword based on the proportion of occurrence of each keyword in the dialogue text of all users of the product recommendation system; calculate the importance of each keyword based on the frequency of occurrence and The importance is used to calculate the recommendation order of each keyword; and the recommended products for each keyword are provided to the user based on the recommendation order. A product recommendation method, including: collecting purchase records, advertisement click records, and browsing records related to at least one product on the Internet for each user; collecting behavior records of each user in a physical store; and obtaining each user's behavior record from the camera and audio recorder of the physical store. The user's images and sounds in the physical store are then analyzed to obtain the user's product interaction records in the physical store. The product interaction records include which items in the physical store the user has viewed. product and the viewing duration exceeds the default value; and Select a target user among these users to provide recommended products to the target user based on the purchase record, the ad click record, the browsing record, the behavior record, and the product interaction record. The product recommendation method described in request item 7, wherein the step of providing recommended products to the target user includes: providing the product to the target user based on the purchase record, the ad click record, the browsing record, the behavior record and the product interaction record. The target user provides the recommended products that users with similar interests to the target user have shown interest in; and based on the purchase record, the ad click record, the browsing record, the behavior record and the product interaction record, provide the target user with The recommended products are similar to the products that the target user has been interested in. A computer-readable medium, used in a computing device or computer, stores instructions to execute the product recommendation method described in any one of claims 6 to 8.

Images