A personalized recommendation online and offline integrated method based on multimodal feature association

Abstract

The present invention relates to the field of computer technology, and discloses an online and offline integrated personalized recommendation method based on multimodal feature association, comprising the following methods: 1) constructing a data acquisition module; 2) constructing a module for extracting multimodal features; 3) constructing a data processing module; 4) constructing a collaborative filtering recommendation algorithm module; 5) constructing an offline and online mapping module for multimodal feature association; 6) constructing a customized field for result filtering module, which combines the current model in natural language processing, performs multimodal feature analysis on text description information contained in online and offline commodities, and through more effective modeling association, can recommend offline commodities to users according to their personalized preferences revealed in their online behaviors even in the absence of offline user shopping data, and can meet some customized parameter screening for the final personalized recommendation of online or offline commodities, thereby solving the problem of being unable to provide users with personalized recommendations for offline commodities.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an online and offline integrated personalized recommendation method based on multimodal feature association. Background Art

[0002] With the development of the times, people have begun to favor the product selection and purchasing methods of e-commerce, which has also brought a great impact on traditional physical stores. The operating costs of physical shopping malls are high, and it is becoming increasingly difficult to operate physical stores. From the customer's perspective, what matters when buying goods is the product itself, not simply the shopping method. Therefore, how to accurately recommend products that meet customer needs is an issue that needs to be addressed urgently.

[0003] Currently, existing technical solutions can complete intelligent product recommendations based on user habits and historical consumption information combined with product information, which can greatly save users' time in searching for products; however, the scope of use is single and only applies to recommendations for online shopping mall products, and cannot help users get more personalized and accurate recommendation results when shopping in physical stores; therefore, how to combine the recommendation methods and personalized information in online e-commerce platforms with the products in offline physical stores to recommend products to users, thereby providing a better shopping experience, is an urgent problem to be solved.

[0004] To this end, we propose an integrated online and offline method for personalized recommendation based on multimodal feature association. Summary of the invention

[0005] The purpose of the present invention is to provide an integrated online and offline method for personalized recommendation based on multimodal feature association, which solves the problem that the background technology has a single scope of use and is only applicable to the recommendation of online shopping mall products, and cannot help users get more personalized and accurate recommendation results when shopping in physical shopping malls.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for integrating online and offline personalized recommendation based on multimodal feature association, comprising the following steps:

[0007] S1: build data acquisition module;

[0008] S2: Construct a module for extracting multimodal features;

[0009] S3: build data processing module;

[0010] S4: Construct collaborative filtering recommendation algorithm module;

[0011] S5: Build an offline and online mapping module for multimodal feature association;

[0012] S6: Build a customized field for result filtering module.

[0013] Furthermore, in S1, the following steps are included:

[0014] S11: Obtain the purchase click history data table of the user on the online platform for the product, and the consumption history table of the user in the offline physical store, including the user code, product code, and behavior time information, and store them in chronological order;

[0015] S12: Obtain store names, store codes, business formats, and location information of all stores;

[0016] S13: Obtain all online product information and offline product information, such as product name, product code, store code, and multimodal content features, such as product text description and product image description.

[0017] Furthermore, S2 also includes:

[0018] S21: Extract all attribute field columns containing text content in the online product information table online_goods, concatenate them, and put them into the pre-trained Bert model to generate corresponding embedded vector representation features for each online product, and save them as a new table online_item_txt_emb; extract the offline product information table offline_goods, do the same operation, and save them as the corresponding new table offline_item_txt_emb.

[0019] Furthermore, S3 includes the following steps:

[0020] S31: Calculate and generate a specified number of popular online products, calculate the number of times each product has been consumed based on the online and offline purchase record data information of the marked users, and count the most popular online popular product set topk_online_click_items of the specified number K;

[0021] S32: Process part of the original data to generate tabular data to be uploaded to the SQL database.

[0022] Further, in S32, the following steps are included:

[0023] S321: Add a rank field column to the original online product attribute table online_goods as the default sorting field, where the rank field column is the frequency of each product in online_goods appearing in online_behaviors as the sorting frequency, and directly fill in the default value 99999 for products that have not appeared. In addition, according to the store information table stores, add the mall code mall_code and store code store_code associated with each product; finally generate the online product table online_goods_for_sql uploaded to the SQL database;

[0024] S322: Add a rank field column to the original offline goods attribute table offline_goods as the default sorting field. First, use faiss to calculate the text content similarity sim_score between each offline product in offline_goods and each popular product in topk_online_click_items.

[0025] S323: By calculating topk_rank*sim_score, where topk_rank is the first K rank fields sorted in descending order by the associated value, the popularity estimation value corresponding to each offline product is obtained, and then sorted;

[0026] S324: According to the store information table, the mall code mall_code and store code store_code associated with each product are added, and finally an offline product table offline_goods_for_sql is generated and uploaded to the SQL database.

[0027] Furthermore, in S4, the following steps are included:

[0028] S41: Build an online product personalized recommendation module to generate online product recommendation results;

[0029] S42: Building an offline store personalized recommendation module to generate offline store recommendation results;

[0030] S43: According to S41 and S42, a personalized recommendation list result_online_list of online products and a personalized recommendation list result_store_list of offline stores are obtained for each user.

[0031] Furthermore, in S41 and S42, when calculating the similarity between two items based on the online consumption behavior table online_behaviors and the offline consumption behavior table offline_behaviors, the interaction time information, direction information, item popularity, and user activity factors are taken into account. For example, when calculating the similarity between item i and item j, the following formula is used:

[0032]

[0033] , where Ui represents the set of users who have interacted with item i; Iu represents the set of items that user u has interacted with, Uj represents the set of users who have interacted with item j, α is the weight factor considering the interaction time difference factor, ti is the interaction time of item i, tj is the interaction time of item j, c is the weight factor considering the item interaction direction factor, β is the parameter factor considering the item interaction direction factor, li is the interaction position of item i, and lj is the interaction position of item j.

[0034] Furthermore, in S5, the following steps are included:

[0035] S51: According to the result_store_list table obtained in S43, the personalized store recommendation list for each user is used as the range, and all the associated offline products are found as the candidate set. Then, according to the result_online_list table, the personalized online product recommendation list corresponding to the user is found. The text similarity between each candidate offline product and the online product is calculated using faiss, and used as the association value of the multimodal feature. After sorting by the size of the association value, the specified number of topK offline products are taken out, and the topK are the first K after being sorted in descending order by the association value.

[0036] S52: In S51, if there is a user who does not have a corresponding online consumption recommendation list in the result_online_list table, the online popular product list topk_online_click_items is used for default filling, and finally the user's offline product personalized recommendation list user_off_items_rec_list_with_rank is obtained;

[0037] S53: According to the result_online_list obtained in S43, it is processed into a key-value pair, and finally the user_off_items_rec_list_with_rank table is obtained. Then, it is convenient to upload it to the redis server for real-time query, and query the request according to the user code in the form of a key-value pair.

[0038] Furthermore, in S6, the following steps are included:

[0039] S61: Upload the two tables obtained in S32 to the MySQL database;

[0040] S62: Upload the two tables obtained in S52 and S53 to the redis server;

[0041] S63: When querying a list of personalized recommended products suitable for a user, first return a list of all recommended product candidates suitable for the user from redis. Then, according to the filter parameters of the real-time query, generate a corresponding SQL query statement, query the corresponding parameter values ​​of the product table in MySQL, filter the products that meet the specified parameter settings from all the recommended candidates, and use them as the final product recommendation for the user, and finally return the query request.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention proposes an online and offline integrated personalized recommendation method based on multimodal feature association. By applying the advantage of online shopping in enabling intelligent personalized recommendation to offline shopping, the method of linking online and offline shopping is adopted. Through the similarity association of multimodal features of image and text, the shopping commodity information corresponding to people's offline shopping is personalized matched and recommended and sorted, and the screening results can be filtered according to customized field attributes, such as region, brand, price, etc., to meet the intelligent shopping needs of users for offline shopping. At the same time, the current model in natural language processing is combined to perform multimodal feature analysis on the text description information contained in online and offline commodities. Through more effective modeling association, offline commodities can be recommended to users according to their personalized preferences revealed in their online behaviors even in the absence of offline user shopping data, and some customized parameter screening of the online or offline commodities recommended for the final personalized recommendation can be met, thereby solving the problem of being unable to provide users with personalized recommendations for offline commodities. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings constituting a part of this application are used to provide a further understanding of this application, so that other features, purposes and advantages of this application become more obvious. The schematic embodiment drawings and their descriptions of this application are used to explain this application and do not constitute an improper limitation on this application. In the drawings:

[0045] Figure 1 This is an overall flow chart of the online and offline integrated method for personalized recommendation based on multimodal feature association of the present invention;

[0046] Figure 2 The S1 flow chart of the online and offline integrated method for personalized recommendation based on multimodal feature association of the present invention;

[0047] Figure 3The S3 flow chart of the online and offline integrated method for personalized recommendation based on multimodal feature association of the present invention;

[0048] Figure 4 The S4 flow chart of the online and offline integrated method for personalized recommendation based on multimodal feature association of the present invention;

[0049] Figure 5 This is the S6 flow chart of the online and offline integrated method for personalized recommendation based on multimodal feature association of the present invention. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0051] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0052] In the present application, the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inner", "outer", "middle", "vertical", "horizontal", "lateral", "longitudinal" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings. These terms are mainly used to better describe the present application and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation.

[0053] In addition, some of the above terms may be used to express other meanings in addition to indicating orientation or positional relationship. For example, the term "on" may also be used to express a certain dependency or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in this application can be understood according to specific circumstances.

[0054] In addition, the term "plurality" shall mean two or more.

[0055] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0056] See also Figure 1 and Figure 2 , a personalized recommendation online and offline integrated method based on multimodal feature association, comprising the following steps:

[0057] S1: build data acquisition module;

[0058] S2: Construct a module for extracting multimodal features;

[0059] S3: build data processing module;

[0060] S4: Construct collaborative filtering recommendation algorithm module;

[0061] S5: Build an offline and online mapping module for multimodal feature association;

[0062] S6: Build a customized field for result filtering module.

[0063] In S1, the following steps are included:

[0064] S11: Obtain the user's purchase click history data table for the product on the online platform and the user's offline physical store consumption history table, including user code, product code, behavior time information, etc., and store them in chronological order;

[0065] S12: Obtain store names, store codes, business formats, and location information of all stores;

[0066] S13: Obtain all online product information and offline product information, such as product name, product code, store code, and multimodal content features, such as product text description, product image description, etc.

[0067] See also Figure 1 and Figure 3 , S2 includes: extracting all attribute field columns containing text content in the online product information table online_goods, concatenating them, and putting them into the pre-trained Bert model to generate corresponding embedded vector representation features for each online product, and saving them as a new table online_item_txt_emb; extracting the offline product information table offline_goods, doing the same operation, and saving them as the corresponding new table offline_item_txt_emb.

[0068] In S3, the following steps are involved:

[0069] S31: Calculate and generate a specified number of popular online products, calculate the number of times each product has been consumed based on the online and offline purchase record data information of the marked users, and count the most popular online popular product set topk_online_click_items of the specified number K;

[0070] S32: Process part of the original data to generate tabular data to be uploaded to the SQL database.

[0071] In S32, the following steps are included:

[0072] S321: Add a rank field column to the original online product attribute table online_goods as the default sorting field, where the rank field column is the frequency of each product in online_goods appearing in online_behaviors as the sorting frequency, and directly fill in the default value 99999 for products that have not appeared. In addition, according to the store information table stores, add the mall code mall_code, store code store_code, etc. associated with each product; finally generate the online product table online_goods_for_sql uploaded to the SQL database;

[0073] S322: Add a rank field column to the original offline goods attribute table offline_goods as the default sorting field. First, use faiss to calculate the text content similarity sim_score between each offline product in offline_goods and each popular product in topk_online_click_items.

[0074] S323: By calculating topk_rank*sim_score, where topk_rank is the first K rank fields sorted in descending order by the associated value, the popularity estimation value corresponding to each offline product is obtained, and then sorted;

[0075] S324: According to the store information table, the mall code mall_code, store code store_code, etc. associated with each product are added, and finally an offline product table offline_goods_for_sql is generated and uploaded to the SQL database.

[0076] See also Figure 4 , in S4, the following steps are included:

[0077] S41: Build an online product personalized recommendation module to generate online product recommendation results;

[0078] S42: Building an offline store personalized recommendation module to generate offline store recommendation results;

[0079] S43: According to S41 and S42, a personalized recommendation list result_online_list of online products and a personalized recommendation list result_store_list of offline stores are obtained for each user.

[0080] In S41 and S42, when calculating the similarity between two items based on the online consumption behavior table online_behaviors and the offline consumption behavior table offline_behaviors, factors such as interaction time information, direction information, item popularity, and user activity are taken into account. For example, when calculating the similarity between item i and item j, the following formula is used:

[0081]

[0082] Among them, Ui represents the set of users who have interacted with item i; Iu represents the set of items interacted with by user u, Uj represents the set of users who have interacted with item j, α is the weight factor considering the interaction time difference factor, ti is the interaction time of item i, tj is the interaction time of item j, c is the weight factor considering the item interaction direction factor, β is the parameter factor considering the item interaction direction factor, li is the interaction position of item i, and lj is the interaction position of item j.

[0083] In S5, the following steps are included:

[0084] S51: According to the result_store_list table obtained in S43, the personalized store recommendation list for each user is used as the range, and all the associated offline products are found as the candidate set. Then, according to the result_online_list table, the personalized online product recommendation list corresponding to the user is found. The text similarity between each candidate offline product and the online product is calculated using faiss, and used as the association value of the multimodal feature. After sorting by the size of the association value, the specified number of topK offline products are taken out, and the topK are the first K after being sorted in descending order by the association value.

[0085] S52: In S51, if there is a user who does not have a corresponding online consumption recommendation list in the result_online_list table, the online popular product list topk_online_click_items is used for default filling, and finally the user's offline product personalized recommendation list user_off_items_rec_list_with_rank is obtained;

[0086] S53: According to the result_online_list obtained in S43, it is processed into a key-value pair, and finally the user_off_items_rec_list_with_rank table is obtained. Then, it is convenient to upload it to the redis server for real-time query, and query the request according to the user code in the form of a key-value pair.

[0087] See also Figure 5 , in S6, the following steps are included:

[0088] S61: Upload the two tables obtained in S32 to the MySQL database;

[0089] S62: Upload the two tables obtained in S52 and S53 to the redis server;

[0090] S63: When querying a list of personalized recommended products suitable for a user, first return a list of all recommended product candidates suitable for the user from redis. Then, according to the filter parameters of the real-time query, generate a corresponding SQL query statement, query the corresponding parameter values ​​of the product table in MySQL, filter the products that meet the specified parameter settings from all the recommended candidates, and use them as the final product recommendation for the user, and finally return the query request.

[0091] In summary: the present invention provides an online and offline integrated personalized recommendation method based on multimodal feature association, comprising: S1: constructing a data acquisition module; S2: constructing a multimodal feature extraction module;

[0092] S3: Build a data processing module; S4: Build a collaborative filtering recommendation algorithm module; S5: Build an offline and online mapping module associated with multimodal features; S6: Build a customized field for result filtering module. At the same time, in S1, it also includes: S11: Obtain the user's purchase and click history data table for the product on the online platform, and the user's offline physical store consumption history table, such as user code, product code, behavior time information, etc., and store them in chronological order; S12: Obtain the store name, store code, business format and location information of all stores; S13: Obtain all online and offline product information, such as product name, product code, store code and multimodal content features, such as product text description, product image description, etc., extract all attribute field columns containing text content in the online product information table online_goods, splice them together, put them into the pre-trained Bert model to generate the corresponding embedded vector representation feature for each online product, and save it as a new table online_item_txt_emb; extract the offline Perform the same operation on the offline_goods product information table and save it as the corresponding new table offline_item_txt_emb. At the same time, calculate and generate a specified number of popular online products. According to the online and offline purchase record data information of the marked users, calculate the number of times each product is consumed, and count the topk_online_click_items of the most popular online popular product set of a specified number K. Process part of the original data to generate table data uploaded to the SQL database. At the same time, add a rank field column to the original online product attribute table online_goods as the default sorting field. The rank field column is the frequency of each product in online_goods appearing in online_behaviors as the sorting frequency. If it has not appeared, fill in the default value 99999 directly. In addition, according to the store information table stores, add the mall code mall_code and store code store_code associated with each product.Finally, the online goods table online_goods_for_sql uploaded to the SQL database is generated, and then the rank field column is added to the original offline goods attribute table offline_goods as the default sorting field. First, each offline product in offline_goods and each popular product in topk_online_click_items are used to calculate the text content similarity sim_score between each other using faiss, and then the popularity estimate corresponding to each offline product is obtained by calculating topk_rank*sim_score, and then sorting is performed. Finally, according to the store information table, the mall code mall_code, store code store_code, etc. associated with each product are added, and finally the offline goods table offline_goods_for_sql uploaded to the SQL database is generated. or_sql, by applying the advantages of online shopping in intelligent personalized recommendation to offline shopping, adopting the linkage mode of online and offline shopping, through the similarity association of multimodal features of image and text, the shopping product information corresponding to people's offline shopping is personalized matched and recommended, and the screening results can be filtered according to customized field attributes, such as region, brand, price, etc., to meet the intelligent shopping needs of users for offline shopping. At the same time, combined with the current model in natural language processing, the text description information contained in online and offline products is multimodal. Through more effective modeling association, even in the absence of offline user shopping data, offline products can still be recommended to users based on their personalized preferences revealed in their online behavior, and some customized parameter screening of the final personalized recommendation of online or offline products can be met, solving the problem of being unable to provide users with personalized recommendations for offline products. ;

[0093] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0094] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A personalized recommendation online and offline integrated method based on multimodal feature association, characterized in that: The following steps are involved: S1: build data acquisition module; S2: Construct a module for extracting multimodal features; S3: build data processing module; S4: Construct collaborative filtering recommendation algorithm module; S5: Build an offline and online mapping module for multimodal feature association; S6: Build customized fields for result filtering module; In S4, the following steps are included: S41: Build an online product personalized recommendation module to generate online product recommendation results; S42: Building an offline store personalized recommendation module to generate offline store recommendation results; S43: According to S41 and S42, a personalized recommendation list result_online_list of online products and a personalized recommendation list result_store_list of offline stores are obtained for each user; In S5, the following steps are included: S51: According to the result_store_list table obtained in S43, the personalized store recommendation list for each user is used as the range, and all the associated offline products are found as the candidate set. Then, according to the result_online_list table, the personalized online product recommendation list corresponding to the user is found. The text similarity between each candidate offline product and the online product is calculated using faiss, and used as the association value of the multimodal feature. After sorting by the size of the association value, the specified number of topK offline products are taken out, and the topK are the first K after being sorted in descending order by the association value. S52: In S51, if there is a user who does not have a corresponding online consumption recommendation list in the result_online_list table, the online popular product list topk_online_click_items is used for default filling, and finally the user's offline product personalized recommendation list user_off_items_rec_list_with_rank is obtained; S53: According to the result_online_list obtained in S43, it is processed into a key-value pair, and finally the user_off_items_rec_list_with_rank table is obtained. Then, it is convenient to upload it to the redis server for real-time query, and query the request according to the user code in the form of a key-value pair.

2. The method for integrated online and offline personalized recommendation based on multimodal feature association according to claim 1, characterized in that: In S1, the following steps are included: S11: Obtain a user's purchase click history data table for products on the online platform and a user's offline physical store consumption history table, which includes user code, product code, behavior time information, and is stored in chronological order; S12: Obtain store names, store codes, business formats, and location information of all stores; S13: Obtain all online product information and offline product information, including product name, product code, store code, and multimodal content features, including product text description and product image description.

3. The method for integrated online and offline personalized recommendation based on multimodal feature association as claimed in claim 2, characterized in that: Also included in S2: S21: Extract all attribute field columns containing text content in the online product information table online_goods, concatenate them, and put them into the pre-trained Bert model to generate corresponding embedded vector representation features for each online product, and save them as a new table online_item_txt_emb; extract the offline product information table offline_goods, do the same operation, and save them as the corresponding new table offline_item_txt_emb.

4. The method for integrated online and offline personalized recommendation based on multimodal feature association as claimed in claim 3, characterized in that: The following steps are involved in S3: S31: Calculate and generate a specified number of popular online products, calculate the number of times each product has been consumed based on the online and offline purchase record data information of the marked users, and count the most popular online popular product set topk_online_click_items of the specified number K; S32: Process part of the original data to generate tabular data to be uploaded to the SQL database.

5. The method for integrated online and offline personalized recommendation based on multimodal feature association as claimed in claim 4, characterized in that: In S32, the following steps are included: S321: Add a rank field column to the original online product attribute table online_goods as the default sorting field, where the rank field column is the frequency of each product in online_goods appearing in online_behaviors as the sorting frequency, and directly fill in the default value 99999 for products that have not appeared. In addition, according to the store information table stores, add the mall code mall_code and store code store_code associated with each product; finally generate the online product table online_goods_for_sql uploaded to the SQL database; S322: Add a rank field column to the original offline goods attribute table offline_goods as the default sorting field. First, use faiss to calculate the text content similarity sim_score between each offline product in offline_goods and each popular product in topk_online_click_items. S323: By calculating topk_rank*sim_score, where topk_rank is the first K rank fields sorted in descending order by the associated value, the popularity estimation value corresponding to each offline product is obtained, and then sorted; S324: According to the store information table, the mall code mall_code and store code store_code associated with each product are added, and finally an offline product table offline_goods_for_sql is generated and uploaded to the SQL database.

6. The method for integrated online and offline personalized recommendation based on multimodal feature association according to claim 1, characterized in that: In S41 and S42, based on the online consumption behavior table online_behaviors and the offline consumption behavior table offline_behaviors, when calculating the similarity between two items, the interaction time information, direction information, item popularity, and user activity factors are taken into account. When calculating the similarity between item i and item j, the following formula is used: Among them, Ui represents the set of users who have interacted with item i; Iu represents the set of items interacted with by user u, Uj represents the set of users who have interacted with item j, α is the weight factor considering the interaction time difference factor, ti is the interaction time of item i, tj is the interaction time of item j, c is the weight factor considering the item interaction direction factor, β is the parameter factor considering the item interaction direction factor, li is the interaction position of item i, and lj is the interaction position of item j.

7. The method for integrated online and offline personalized recommendation based on multimodal feature association according to claim 1, characterized in that: In S6, the following steps are included: S61: Upload the two tables obtained in S32 to the MySQL database; S62: Upload the two tables obtained in S52 and S53 to the redis server; S63: When querying a list of personalized recommended products suitable for a user, first return the list of all recommended product candidates suitable for the user from redis, and then generate the corresponding SQL query statement based on the filter parameters of the real-time query. According to the parameter values ​​corresponding to the product table in MySQL, filter the products that meet the specified parameter settings from all the recommended candidates as the final product recommendation for the user, and finally return the query request.

Images