A recommendation method and device, electronic equipment and readable storage medium

Abstract

The present disclosure provides a recommendation method, device, electronic equipment and readable storage medium, comprising: obtaining access data of the user; the access data comprises historical access data and / or real-time access data; inputting the access data into a pre-trained machine learning model to generate a recommended merchant commodity list for the user; the merchant commodity list comprises recommended merchants and recommended commodities for at least one recommended merchant; and the merchants and commodities in the recommended merchant commodity list are correspondingly displayed to the user. The problem that the recommendation method in the prior art cannot actually combine the comprehensive needs of the user, resulting in the fact that the recommendation cannot meet the user's desire to obtain comprehensive information of commodities and merchants.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of recommendation technology, and in particular to a recommendation method, device, electronic device, and readable storage medium. Background Art

[0002] With the widespread use of mobile terminals in people's lives, people are generally accustomed to using mobile applications to obtain life information. Among them, offline mining of user historical behavior data to understand user interests and recommend products that meet their needs is a more common method.

[0003] In existing technology, recommendation systems simply recommend merchants or products based on a user's historical information. For example, in a takeout scenario, users need recommendations not only for merchants but also for dishes they might be interested in. Therefore, a recommendation system determines whether a user requires a package deal based on the user's historical order data. This system not only recommends merchants of interest but also recommends products the user might like. This shortens the user's decision-making process, reduces ordering costs, and increases their desire to order.

[0004] Therefore, existing technologies cannot meet users' needs for comprehensive information acquisition on products and merchants. Summary of the Invention

[0005] In view of the above problems, the embodiments of the present disclosure are proposed to provide a recommendation method and a corresponding recommendation device that overcome the above problems or at least partially solve the above problems.

[0006] According to the first aspect of the present disclosure, an embodiment of the present disclosure discloses a recommendation method, which specifically includes:

[0007] Acquiring access data of the user; the access data includes historical access data and / or real-time access data;

[0008] Inputting the access data into a pre-trained machine learning model to generate a recommended merchant product list for the user; the merchant product list includes recommended merchants and recommended products for at least one recommended merchant;

[0009] The merchants and products in the recommended merchant product list are displayed to the user accordingly.

[0010] According to a second aspect of the present disclosure, an embodiment of the present disclosure discloses a recommendation device, specifically comprising:

[0011] A data acquisition module, configured to acquire access data of the user; the access data includes historical access data and / or real-time access data;

[0012] a recommended merchant product list generation module, configured to input the access data into a pre-trained machine learning model to generate a recommended merchant product list for the user; the merchant product list includes recommended merchants and recommended products for at least one of the recommended merchants;

[0013] The recommended merchant product list display module is used to display the merchants and products in the recommended merchant product list to the user accordingly.

[0014] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the aforementioned recommendation method when executing the program.

[0015] According to a fourth aspect of the present disclosure, a readable storage medium is provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device can implement the aforementioned recommendation method.

[0016] The embodiments of the present disclosure include the following advantages: by obtaining the access data of the user; the access data includes historical access data and / or real-time access data; the access data is input into a pre-trained machine learning model to generate a recommended merchant product list for the user; the merchant product list includes recommended merchants and recommended products for at least one recommended merchant; the merchants and products in the recommended merchant product list are displayed to the user accordingly, providing the user with a method for recommending comprehensive merchant product information. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the description of the embodiments of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0018] Figure 1 is a flowchart of the steps of a recommended method embodiment of the present disclosure;

[0019] Figure 2 is a flowchart of the steps of a recommended method embodiment of the present disclosure;

[0020] Figure 3 is a structural block diagram of a recommended device embodiment of the present disclosure;

[0021] Figure 4 It is a structural block diagram of an embodiment of a recommendation device disclosed in the present invention. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present disclosure in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present disclosure without making any creative efforts shall fall within the scope of protection of the present disclosure.

[0023] Example 1

[0024] Reference Figure 1 , shows a flowchart of a recommended method embodiment of the present disclosure, which may specifically include the following steps:

[0025] Step 101: Acquire the user's access data; the access data includes historical access data and / or real-time access data;

[0026] The embodiments of the present disclosure are applicable to application scenarios in which a recommendation list is displayed through an application on a mobile terminal. The application on the mobile terminal can be any application that provides a product recommendation function, and the embodiments of the present disclosure are not limited to the specific application. The mobile terminal can be any mobile terminal such as a smartphone, a tablet computer, a laptop computer, etc., and the embodiments of the present disclosure are not limited to the specific mobile terminal. For ease of description, the embodiments of the present disclosure are mainly explained using a smartphone as an example, and the application scenarios of other mobile terminals can be referenced with each other.

[0027] When a user obtains a list of recommendations through an application on a mobile terminal, they first trigger a recommendation list request by clicking a virtual button or a similar function key on the mobile terminal application interface. This triggers the backend to send the recommendation list request to the backend. The backend then receives the recommendation list request corresponding to the user based on the target product selected by the mobile terminal user when sending the recommendation request and the relevant target product attributes. Of course, the method for sending and activating the recommendation list request can also be activated when the terminal application is started. The embodiment of this disclosure does not limit the method for activating the recommendation list request.

[0028] Specifically, when a recommendation request is activated on the client, relevant access data is obtained according to the recommendation request. For example, for a dish recommendation request, access data of all dishes is obtained. The access data includes historical access data and / or real-time access data.

[0029] For example, after receiving the recommendation list request, the backend server obtains the historical access data of the user who sent the request, as well as the real-time food-related access data generated when the user is performing application operations, such as browsing and clicking operations on the current food merchant.

[0030] Specifically, the historical access data includes user attribute data, merchant attribute data, user historical visit behavior data of merchants, product attribute data, and user historical visit behavior data of products.

[0031] Real-time access data includes real-time user access to merchants (e.g., merchant attribute information that users are currently browsing and clicking on) and real-time user access to products (e.g., product attribute information that users are currently browsing and clicking on).

[0032] Step 102: Input the access data into a pre-trained machine learning model to generate a recommended merchant product list for the user; the merchant product list includes recommended merchants and recommended products for at least one recommended merchant;

[0033] In an embodiment of the present disclosure, the access data obtained above is input into a pre-trained machine learning model to generate a recommended merchant product list for the user, which includes merchant product recommendation information and is displayed to the user in the form of a list.

[0034] Step 103: Display the merchants and products in the recommended merchant product list to the user accordingly.

[0035] In the embodiment of the present disclosure, after obtaining the recommended merchant product list corresponding to the corresponding user's recommendation list request, the list can be returned to the user and displayed in the application interface for further operation by the user. The embodiment of the present disclosure does not limit the display format of the recommendation request data pushed to the user.

[0036] Among them, when displaying merchant product information recommended to users, the corresponding products can be displayed mainly with the merchant, or mainly with the product and merchant information attached, which can be determined by the user's display preference.

[0037] In an embodiment of the present disclosure, the user's access data is obtained; the access data includes historical access data and / or real-time access data; the access data is input into a pre-trained machine learning model to generate a recommended merchant product list for the user; the merchant product list includes recommended merchants and recommended products for at least one recommended merchant; the merchants and products in the recommended merchant product list are displayed to the user. This solves the problem in the prior art that the recommendation method cannot actually combine the comprehensive needs of users, resulting in the recommendation being unable to meet the user's desire to obtain comprehensive information about products and merchants. It achieves the purpose of meeting the user's need to obtain comprehensive information about products and merchants.

[0038] Example 2

[0039] Reference Figure 2 , shows a flowchart of a recommended method embodiment of the present disclosure, which may specifically include the following steps:

[0040] Step 201: Acquire the user's access data; the access data includes historical access data and / or real-time access data;

[0041] This step is the same as step 101 and will not be described in detail here.

[0042] Step 202: Perform feature conversion on the historical access data to obtain a first feature vector of a preset dimension; the historical access data includes one or more of user attribute data, merchant attribute data, user historical merchant visit behavior data, product attribute data, and user historical product visit behavior data;

[0043] Specifically, after obtaining the historical access data, the user attribute data, merchant attribute data, user historical merchant access behavior data, product attribute data, and user historical product access behavior data contained in the historical access data are converted into feature vectors and then spliced ​​to obtain a first feature vector of a preset dimension.

[0044] The feature vector conversion can be performed using common feature vector conversion tools such as word2vec or one-hot. If the historical access data contains five items of data, each item of data is converted to 25 dimensions, and finally the vectors are concatenated into a 125-dimensional first feature vector.

[0045] It can be understood that the feature vector conversion method is not limited to the above description, and the embodiments of the present disclosure are not limited thereto.

[0046] Step 203 , performing feature conversion on the real-time access data to obtain a second feature vector of a preset dimension; the real-time access data includes real-time user access behavior data of merchants and real-time user access behavior data of products.

[0047] Similarly, the real-time access data is converted into a second feature vector of a preset dimension using the same feature conversion method.

[0048] It can be understood that attribute information is the main information that defines the identity and specific content of users, merchants, and products. It is not limited to the above description. The attribute information is different for merchants and products of different categories, as well as user operation types. Therefore, the embodiments of this disclosure do not limit the specific content of the attribute information.

[0049] Step 204: Input the first feature vector into a pre-trained first machine learning model to obtain the user's historical merchant scores and historical product scores;

[0050] In the disclosed embodiment, based on the user's access operations to different merchants and products in the above historical access data, the merchants and products are ranked according to different scores, and a candidate merchant product ranking list is generated.

[0051] Of course, when generating a sorted list, you can also make a judgment based on the distance between each merchant and product and the user's location. For example, with the user's current location as the center, set a higher score for merchants and all products within a five-kilometer circle. Secondly, the user's historical visit behavior to all merchants, such as placing orders with higher scores, and the scores of comments, clicks, browsing and other operations decrease in turn. Finally, a score is obtained for each merchant, and each product provided by each merchant also gets a score. Merchants and products are sorted according to the scores of the above merchants and products to obtain a candidate merchant product sorting list.

[0052] Among them, the above-mentioned five-kilometer circle with the user's current location as the center is a preset range based on the user's current location. The range is set by relevant technical personnel and can also be manually set by the user according to his current actual situation. This disclosure does not limit this.

[0053] It is understandable that if there is no user history access data within the preset range of the user's current location, a default candidate merchant product ranking list is generated based on the distances between all merchants and the user within the preset range.

[0054] It can be understood that the generation of the default candidate merchant product ranking list is not limited to the distance described above, and the default ranking can also be performed according to the merchant attributes selected by the user. Taking food recommendation as an example, the user may choose Sichuan cuisine, then all merchants and products with Sichuan cuisine attributes will receive a higher score, and a default candidate merchant product ranking list will be generated based on the score.

[0055] Specifically, the first feature vector is input into a preset second machine learning model to obtain the user's historical visit merchant scores and product scores.

[0056] Among them, the preset second machine learning model is trained based on the historical access data of all registered users pre-stored in the system. During training, it is usually necessary to mark the scores of different user access behaviors and labels in the historical access data, such as click count scores, access time scores, etc., and then convert the feature vectors through the marked data and train it through the machine learning algorithm.

[0057] The product score results output by the preset second machine learning model are shown in Table 1:

[0058] state Click No click <![CDATA[(user1,product1)]]> 1 3 <![CDATA[(user2,product2)]]> 2 4 <![CDATA[(user i ,product j )]]> -100 1 … … …

[0059] Table 1

[0060] Among them, user i Combine features for users, product j For products, the status is divided into two types: user clicked and not clicked. Usually, different reward scores are set for different user operation statuses. The user's initial score is 20, and then decays to 1 after 1 minute (calculated based on the distance returned to the sorting result, that is, the faster the user clicks, the higher the score). If the user does not click, a reward of -1000 is given. If the order is weakly successful, a reward of 50 is given. This is deduced in this way to obtain the scores of all merchants or products.

[0061] It can be understood that the above-mentioned scores are calculated based on user and product features, and also include merchant feature scores in the same form. Status can also be added for products and merchants, such as browsing time, number of orders, and number of reviews to set scores. Therefore, the embodiments of the present disclosure do not limit the way of setting scores and the presentation form of the training model for merchants and products.

[0062] Specifically, the first feature vector of the preset dimension obtained above is input into the first machine learning model to obtain the historical merchant score.

[0063] Among them, the first machine learning model is obtained by training based on the user attribute data, the merchant attribute data, and the user's historical visit behavior data of the merchant in the historical access data of all registered users pre-stored in the system. Usually, during training, it is necessary to mark the user's historical visit behavior data of the merchant, such as the number of clicks score, the visit time score, etc., and then convert the feature vector through the marked data, and obtain it through training through the machine learning algorithm.

[0064] Similarly, the output method of the first machine learning model is still as shown in Table 1. The output content includes users and merchants. Table 1 calculates scores based on user and merchant features. The output result is a score calculated based on the features of users and merchants. In addition, status can be added for merchants, such as browsing time, number of orders, and number of evaluations to set scores. Therefore, the embodiment of the present disclosure does not limit the way to set scores and the presentation form of the training model for merchants.

[0065] Specifically, according to the comprehensive scores of merchants and products obtained above, after scoring the merchants and products within a preset range for the current user in the historical data, a ranked list of candidate merchants and products is determined according to the scores.

[0066] Step 205: Input the second feature vector into a pre-trained second machine learning model to obtain the user's real-time merchant score and real-time product score;

[0067] Specifically, the second feature vector obtained above is input into a pre-trained second machine learning model. This second machine learning model uses the same training method as the first, but with different training data: real-time access data. By assigning scores to the real-time access data and training a second machine learning model, the currently collected access data is fed into the second machine learning model to output the current merchant and product scores.

[0068] Step 206: Sort the user's historically visited merchants and historically visited products according to the historical merchant scores and historical product scores, and generate a historical merchant ranking list and a historical product ranking list;

[0069] Specifically, after obtaining the scores of merchants through the first machine learning model, the merchants are sorted according to the scores to obtain a historical merchant sorting list.

[0070] Preferably, it further comprises:

[0071] Sub-step 2061 , based on the user's real-time visit to merchant behavior data and the user's real-time visit to commodity behavior data, respectively update the historical merchant ranking list and the historical commodity ranking list.

[0072] Preferably, the user's real-time merchant access behavior data includes real-time merchant click-through rate, the real-time product access behavior data includes real-time product click-through rate, and the real-time product access behavior data includes real-time product click-through rate.

[0073] Then sub-step 2061 further includes:

[0074] Sub-step 20611: updating the historical merchant ranking list according to the real-time merchant click-through rate;

[0075] Sub-step 20612: updating the historical merchant ranking list according to the real-time product click-through rate.

[0076] Among them, if the user's real-time merchant access behavior data includes real-time merchant click-through rate, then the above-mentioned historical merchant ranking list is updated according to the real-time merchant click-through rate. For example, the historical merchant ranking list has a total of m*n rows, representing m*n states, and the action corresponding to each state has a utility value. The user's real-time data corresponds to different reward scores for the decision of clicking or not clicking in the above list. The sorting algorithm will first find the corresponding row based on the current sorting result, and then compare the values ​​of the two columns (click and not click), select the action with the larger value as the current action, and finally traverse all rows and columns, and update the score corresponding to each state according to different actions. The final result is the candidate merchant ranking list.

[0077] Step 207: Sort the real-time merchants and real-time products visited by the user according to the real-time merchant scores and the real-time product scores, and generate a real-time merchant ranking list and a real-time product ranking list;

[0078] According to the above-obtained real-time merchant scores and real-time commodity scores, a real-time merchant ranking list and a real-time commodity ranking list are generated and displayed in order according to the scores.

[0079] Step 208: Based on the historical merchant ranking list and the historical product ranking list, the real-time merchant ranking list and the real-time product ranking list are respectively modified to generate a recommended merchant product list for the user; the merchant product list includes recommended merchants and recommended products for at least one recommended merchant;

[0080] Preferably, step 208 further includes:

[0081] Sub-step 2081, performing weighted summation on the historical merchant ranking list and product weights, and the historical product ranking list and merchant weights, to obtain a merchant product weight ranking list;

[0082] Sub-step 2082: According to the order of merchant products in the merchant product weight ranking list, the real-time merchant ranking list and the real-time product ranking list are modified to generate a recommended merchant product list for the user.

[0083] It can be understood that historical access data is the historical operation data generated by the user in the current application within a preset time period, including user attributes (such as gender, age, occupation, usual location, taste preferences, dish preferences, merchant preferences, etc.), merchant attribute information (such as merchant brand information, location information, comprehensive evaluation information, business hours information, provided dish information, etc. within the specified range of the user's location), product attribute information (all provided food information within the user's location, food location information, sales, taste, evaluation, etc.), user's historical visit to merchant behavior data (such as merchant information that users have visited, ordered, browsed, and evaluated, which can be marked with different scores for each behavior), user's historical visit to product behavior data (such as product information that users have ordered, browsed, and evaluated, which can be marked with different scores for different behaviors).

[0084] Therefore, the user preference errors stored in the generated real-time merchant ranking list and real-time product ranking list can be corrected using the user's historical access data, that is, the merchant and product rankings in the real-time merchant ranking list and real-time product ranking list can be corrected based on the merchant product order in the merchant product weight ranking list obtained according to the historical access data.

[0085] Specifically, the product weight and merchant weight are set by relevant technical personnel. Usually, users place orders for specific products, so the merchant weight is usually greater than the product weight. The candidate merchant ranking list and the product weight, as well as the candidate product ranking list and the merchant weight are weighted and summed. That is, each status score in the candidate merchant ranking list and the candidate product ranking list is multiplied by the weight value to obtain the merchant product weight ranking list.

[0086] Step 209: Display the merchants and products in the recommended merchant product list to the user accordingly.

[0087] This step is the same as step 107 and will not be described in detail here.

[0088] Preferably, the first machine learning model is trained by the following steps:

[0089] Step A1, extracting attribute data of each user, attribute data of each merchant, and visit rate of each user to each merchant within a preset time period;

[0090] In the disclosed embodiment, access data for merchants or products by all users within a preset time period are extracted, including attribute data of each user (such as gender, age, occupation, usual location, taste preference, dish preference, merchant preference, etc.), attribute data of each merchant (such as brand information, location information, comprehensive evaluation information, business hours information, provided dish information, etc. of merchants within a specified range of the user's location), and access rate of each user to merchants (such as marking merchant information that users have visited, ordered, browsed, and evaluated within the preset time period).

[0091] Step A2, scoring each merchant according to the visit rate of each user to each merchant, and obtaining characteristic data of each user's visit to each merchant;

[0092] In the embodiment of the present disclosure, different visit behaviors of users to merchants in the above data are marked with scores, that is, scored, to obtain characteristic data of each user's visit to the merchant.

[0093] Step A3, performing feature conversion on the attribute data of each user, the attribute data of each merchant, and the characteristic data of each user's visit to each merchant to obtain a second training feature vector of a preset dimension;

[0094] As shown in Table 1, feature conversion is performed on the attribute data of each user after scoring each behavior state, the attribute data of each merchant, and the characteristic data of each user's visit to each merchant to obtain a second training feature vector of a preset dimension.

[0095] Step A4, extracting attribute data of each user, attribute data of each product, and visit rate of each user to the product within a preset time period;

[0096] In the disclosed embodiment, access data for merchants or products by all users within a preset time period are extracted, including attribute data of each user (such as gender, age, occupation, usual location, taste preference, dish preference, merchant preference, etc.), attribute data of each product (all provided food information within the user's location, food location information, sales volume, taste, evaluation, etc.), and the access rate of each user to the product (such as marking the product information that the user has ordered, browsed, and evaluated within the preset time period).

[0097] Step A5: scoring each product according to the visit rate of each user to each product, and obtaining characteristic data of each user's visit to each product;

[0098] In the embodiment of the present disclosure, different access behaviors of users to commodities in the above data are marked with scores, that is, scored, to obtain characteristic data of each user's access to commodities.

[0099] Step A6: performing feature conversion on the user attribute data, the product attribute data, and the user's access feature data for each product to obtain a product training feature vector of a preset dimension;

[0100] Specifically, after obtaining the above-mentioned initial score setting table (as shown in the display form of Table 1), initial data for the user's product access status score is generated, and the data is converted into a product training feature vector of a preset dimension.

[0101] Step A7: Train the second training feature vector and the product training feature vector respectively using a preset machine learning algorithm to obtain a first machine learning model.

[0102] In the disclosed embodiment, the third training feature vector is trained using a preset machine learning algorithm to obtain a second machine learning model.

[0103] Preferably, the second machine learning model is trained by the following steps:

[0104] Step B1: extracting access data of all users to merchants or products within a preset time period, wherein the access data includes attribute data of each user, attribute data of each merchant, attribute data of each product, access rate of each user to the merchant, and access rate of each user to the product;

[0105] In the embodiment of the present disclosure, access data of all users to merchants or products within a preset time period are extracted, including attribute data of each user (such as gender, age, occupation, usual location, taste preference, dish preference, merchant preference, etc.), attribute data of each merchant (such as merchant brand information, location information, comprehensive evaluation information, business hours information, provided dish information, etc. within the specified range of the user's location), attribute data of each product (all provided food information within the user's location, food location information, sales volume, taste, evaluation, etc.), the access rate of each user to the merchant (such as marking the merchant information that the user has visited, ordered, browsed, and evaluated within the preset time period), and the access rate of each user to the product (such as marking the product information that the user has ordered, browsed, and evaluated within the preset time period).

[0106] Step B2: scoring each merchant based on each user's visit behavior data to the merchant, and obtaining characteristic data of each user's visit to the merchant;

[0107] Step B3, scoring each product based on each user's access behavior data to obtain characteristic data of each user's access to the product;

[0108] For example, using the Q-leaning algorithm, Q is the action utility function, which is used to evaluate the pros and cons of taking a certain action under a specific state. It can be understood as the brain of the sorting algorithm (Agent). We can think of Q as a table. Each row in the table represents a state, and each column (there are two columns in this problem) represents an action (click or not click). This table has a total of m*n rows, representing m*n states, and the action corresponding to each state has a utility value. After training, the user's decision to click or not click in the recommendation list is determined by this table. The sorting algorithm will first find the corresponding row based on the current sorting result, then compare the values ​​of the two columns (click or not click), and select the action with the larger value as the current action.

[0109] Step B4, performing feature conversion on the user attribute data, the merchant attribute data, the product attribute data, the characteristic data of each user's visit to the merchant, and the characteristic data of each user's visit to the product to generate a first training feature vector of a preset dimension;

[0110] Specifically, after obtaining the above-mentioned initial score setting table (as shown in the display form of Table 1), initial data for user merchant, product, and access status scores are generated, and the data is converted into a first training feature vector of a preset dimension.

[0111] Step B5: Train the first training feature vector using a preset machine learning algorithm to obtain a second machine learning model.

[0112] Specifically, the training steps are described as follows:

[0113] A, initialize the above Q value;

[0114] B, initialize the initial sorting state S;

[0115] C, based on Q and S, use a strategy to get action A (e.g., multiple strategies are randomly selected, i.e., the mapping relationship between different actions and scores does not make each iteration proceed along the highest Q value);

[0116] D performs action A (e.g., click, browse, place an order, etc.), obtains a new ranking result (e.g., updates the corresponding status column score according to the new action), obtains a new state S′, and receives a reward R, where the reward can be 1, 50, or -1000;

[0117] E, mathematical description such as: Q(S, A)←(1-α)*Q(S, A)+α*[R+γ*maxQ(S', a)] updates S in Q; where α is the learning rate and γ is the discount factor; it can be seen that the larger γ is, the more the sorting algorithm will value past experience, and the smaller it is, the sorting algorithm will only value immediate interests (R).

[0118] F, S←S′, updates the old sorting state by obtaining the new sorting state.

[0119] G. Repeat the operations from step 2 until the scores corresponding to all actions are added up.

[0120] Among them, the mapping strategy between actions and scores obtained by the above method is the training parameter, and the training of the second machine learning model is completed.

[0121] In an embodiment of the present disclosure, access data of the user is obtained; the historical access data is subjected to feature conversion to obtain a first feature vector of a preset dimension; the historical access data includes one or more of user attribute data, merchant attribute data, historical merchant behavior data of the user, product attribute data, and historical product behavior data of the user; the real-time access data is subjected to feature conversion to obtain a second feature vector of a preset dimension; the first feature vector is input into a pre-trained first machine learning model to obtain the user's historical merchant score and historical product score respectively; the second feature vector is input into a pre-trained second machine learning model to obtain the user's real-time merchant score and real-time product score; the user's historically visited merchants and historically visited products are sorted according to the historical merchant scores and the historical product scores to generate a historical merchant sorting list and a historical product sorting list; the user's real-time visited merchants and real-time visited products are sorted according to the real-time merchant scores and the real-time product scores to generate a real-time merchant sorting list and a real-time product sorting list; the real-time merchant sorting list and the real-time product sorting list are respectively modified according to the historical merchant sorting list and the historical product sorting list to generate a recommended merchant product list for the user. The merchant product list includes recommended merchants and recommended products for at least one recommended merchant; the merchants and products in the recommended merchant product list are displayed to the user accordingly, providing the user with a comprehensive merchant product information recommendation method, meeting the user's need to obtain comprehensive recommendation information of product merchants.

[0122] Example 3

[0123] Reference Figure 3 , shows a structural block diagram of an embodiment of a recommendation device disclosed herein, which may specifically include the following modules:

[0124] The data acquisition module 301 is used to acquire the user's access data; the access data includes historical access data and / or real-time access data;

[0125] A recommended merchant product list generation module 302 is configured to input the access data into a pre-trained machine learning model to generate a recommended merchant product list for the user; the merchant product list includes recommended merchants and recommended products for at least one of the recommended merchants;

[0126] The recommended merchant product list display module 303 is used to display the merchants and products in the recommended merchant product list to the user accordingly.

[0127] In the disclosed embodiment, a data acquisition module is used to acquire the user's access data; the access data includes historical access data and / or real-time access data; a recommended merchant product list generation module is used to input the access data into a pre-trained machine learning model to generate a recommended merchant product list for the user; the merchant product list includes recommended merchants and recommended products for at least one recommended merchant; and a recommended merchant product list display module is used to display the merchants and products in the recommended merchant product list to the user. This achieves the purpose of satisfying the user's need for comprehensive information acquisition on products and merchants.

[0128] Example 4

[0129] Reference Figure 4 , shows a structural block diagram of an embodiment of a recommendation device disclosed herein, which may specifically include the following modules:

[0130] The data acquisition module 401 is used to acquire the user's access data; the access data includes historical access data and / or real-time access data;

[0131] A recommended merchant product list generation module 402 is configured to input the access data into a pre-trained machine learning model to generate a recommended merchant product list for the user; the merchant product list includes recommended merchants and recommended products for at least one of the recommended merchants;

[0132] Preferably, the recommended merchant product list generating module 402 further includes:

[0133] Feature conversion submodule 4021 is used to perform feature conversion on the historical access data to obtain a first feature vector of a preset dimension; the historical access data includes one or more of user attribute data, merchant attribute data, user historical merchant visit behavior data, product attribute data, and user historical product visit behavior data;

[0134] A second feature vector acquisition submodule 4022 is configured to perform feature conversion on the real-time access data to obtain a second feature vector of a preset dimension;

[0135] Preferably, the real-time access data includes real-time user access behavior data of merchants and real-time user access behavior data of products.

[0136] A historical score acquisition submodule 4023 is configured to input the first feature vector into a pre-trained first machine learning model to obtain the user's historical merchant scores and historical product scores;

[0137] A real-time score submodule 4024 is configured to input the second feature vector into a pre-trained second machine learning model to obtain the user's real-time merchant score and real-time product score;

[0138] A historical ranking list generation submodule 4025 is configured to rank the user's historically visited merchants and historically visited commodities according to the historical merchant scores and historical commodity scores, and generate a historical merchant ranking list and a historical commodity ranking list;

[0139] The historical ranking list updating submodule is used to update the historical merchant ranking list and the historical product ranking list respectively according to the user's real-time access to merchant behavior data and the user's real-time access to product behavior data.

[0140] Preferably, the user's real-time merchant access behavior data includes a real-time merchant click-through rate, the real-time product access behavior data includes a real-time product click-through rate, and the real-time product access behavior data includes a real-time product click-through rate, then the historical sort list updating submodule further includes:

[0141] A historical merchant ranking list updating unit, configured to update the historical merchant ranking list according to the click-through rate of the real-time merchant;

[0142] The historical merchant ranking list updating unit is used to update the historical merchant ranking list according to the real-time commodity click-through rate.

[0143] A real-time ranking list generating submodule 4026 is configured to rank the real-time merchants and real-time commodities visited by the user according to the real-time merchant scores and the real-time commodity scores, and generate a real-time merchant ranking list and a real-time commodity ranking list;

[0144] The recommended merchant product list generation submodule 4027 is used to modify the real-time merchant ranking list and the real-time product ranking list according to the historical merchant ranking list and the historical product ranking list, and generate a recommended merchant product list for the user.

[0145] Preferably, the recommended merchant product list generation submodule 4027 further includes:

[0146] Perform weighted summation on the historical merchant ranking list and product weights, and on the historical product ranking list and merchant weights to obtain a merchant product weight ranking list;

[0147] According to the order of merchant products in the merchant product weight ranking list, the real-time merchant ranking list and the real-time product ranking list are modified to generate a recommended merchant product list for the user.

[0148] The recommended merchant product list display module 403 is used to display the merchants and products in the recommended merchant product list to the user accordingly.

[0149] Preferably, the historical score acquisition submodule 4023 further includes:

[0150] The second machine learning model training module is used to train the first machine learning model.

[0151] The second machine learning model training module further includes:

[0152] The merchant visit rate extraction submodule is used to extract the attribute data of each user, the attribute data of each merchant, and the visit rate of each user to each merchant within a preset time period;

[0153] The access characteristic data obtaining submodule is used to score each merchant according to the access rate of each user to each merchant, and obtain the access characteristic data of each user to each merchant;

[0154] A second training feature vector obtaining submodule is configured to perform feature conversion on the attribute data of each user, the attribute data of each merchant, and the characteristic data of each user's visit to each merchant to obtain a second training feature vector of a preset dimension;

[0155] The product visit rate extraction submodule is used to extract the attribute data of each user, the attribute data of each product, and the visit rate of each user to the product within a preset time period;

[0156] The access characteristic data obtaining submodule is used to score each commodity according to the access rate of each user to each commodity, and obtain the access characteristic data of each user for each commodity;

[0157] A product training feature vector acquisition submodule is used to perform feature conversion on the attribute data of each user, the attribute data of each product, and the access feature data of each user to each product to obtain a product training feature vector of a preset dimension;

[0158] The first machine learning model obtaining submodule is used to train the second training feature vector and the product training feature vector respectively through a preset machine learning algorithm to obtain a first machine learning model.

[0159] Preferably, the real-time scoring submodule 4024 further includes:

[0160] The second model training module is used to train the second machine learning model.

[0161] Wherein, the second model training module further includes:

[0162] The access rate extraction submodule is used to extract the access data of all users to merchants or products within a preset time period. The access data includes the attribute data of each user, the attribute data of each merchant, the attribute data of each product, the access rate of each user to the merchant, and the access rate of each user to the product;

[0163] A submodule for obtaining characteristic data of users' visits to merchants, which is used to score each merchant based on the visit behavior data of each user to the merchant, and obtain the characteristic data of each user's visits to the merchant;

[0164] A submodule for obtaining characteristic data of user access to commodities, which is used to score each commodity based on the access behavior data of each user to the commodity, and obtain characteristic data of each user access to the commodity;

[0165] a submodule for generating a first training feature vector of a preset dimension, configured to perform feature conversion on the user attribute data, the merchant attribute data, the product attribute data, the characteristic data of each user's visit to the merchant, and the characteristic data of each user's visit to the product, to generate a first training feature vector of a preset dimension;

[0166] The second machine learning model obtaining submodule is used to train the first training feature vector through a preset machine learning algorithm to obtain a second machine learning model.

[0167] In the embodiment of the present disclosure, a data acquisition module is used to obtain the access data of the user; the access data includes historical access data and / or real-time access data; a recommended merchant product list generation module is used to input the access data into a pre-trained machine learning model to generate a recommended merchant product list for the user; the merchant product list includes recommended merchants and recommended products for at least one recommended merchant; a recommended merchant product list display module is used to display the merchants and products in the recommended merchant product list to the user. The purpose of satisfying the user's need for comprehensive information acquisition on products and merchants is achieved. In addition, by training the first machine learning model and the second machine learning model, behavior and score marking is completed through user data, merchant data, product data, user behavior data on merchants, and user behavior data on products, and finally the model training is completed to achieve the purpose of scoring and sorting the predicted data.

[0168] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0169] An embodiment of the present disclosure further provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements one or more of the above-described recommendation methods when executing the program.

[0170] An embodiment of the present disclosure further provides a readable storage medium. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the recommendation method as described above.

[0171] In summary, the embodiments of the present disclosure have the following advantages:

[0172] 1. Provide users with a way to recommend merchant product combinations;

[0173] 2. Multi-model combination (merchant model + product model + merchant and product model) improves recommendation effect and recommendation experience.

[0174] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0175] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, devices, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0176] The embodiments of the present disclosure are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0177] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0178] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0179] Although the preferred embodiments of the present disclosure have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic creative concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present disclosure.

[0180] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0181] The above is a detailed introduction to a recommended method, device, electronic device and readable storage medium provided by the present disclosure. Specific examples are used herein to illustrate the principles and implementation methods of the present disclosure. The description of the above embodiments is only used to help understand the method and core ideas of the present disclosure. At the same time, for those skilled in the art, according to the ideas of the present disclosure, there may be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present disclosure.

Claims

1. A recommendation method, characterized in that: include: Obtaining user access data; the access data includes historical access data and / or real-time access data; Input the access data into a pre-trained machine learning model to generate a list of recommended merchant products for the user, including: Performing feature conversion on the historical access data to obtain a first feature vector of a preset dimension; the historical access data includes user attribute data, merchant attribute data, user historical visit behavior data of merchants, product attribute data, and user historical visit behavior data of products; Performing feature conversion on the real-time access data to obtain a second feature vector of a preset dimension; Inputting the first feature vector into a pre-trained first machine learning model to obtain the user's historical merchant scores and historical product scores respectively; Inputting the second feature vector into a pre-trained second machine learning model to obtain the user's real-time merchant score and real-time product score; Sort the user's historically visited merchants and historically visited commodities according to the historical merchant scores and the historical commodity scores, and generate a historical merchant sorting list and a historical commodity sorting list; Sort the real-time merchants and real-time commodities visited by the user according to the real-time merchant scores and the real-time commodity scores, and generate a real-time merchant ranking list and a real-time commodity ranking list; According to the historical merchant ranking list and the historical product ranking list, respectively modifying the real-time merchant ranking list and the real-time product ranking list to generate a recommended merchant product list for the user, including: Perform weighted summation on the historical merchant ranking list and product weights, and on the historical product ranking list and merchant weights to obtain a merchant product weight ranking list; According to the merchant product order in the merchant product weight ranking list, the real-time merchant ranking list and the real-time product ranking list are modified to generate a recommended merchant product list for the user; The merchant product list includes recommended merchants and recommended products for the recommended merchants; The merchants and products in the recommended merchant product list are displayed to the user accordingly.

2. The method according to claim 1, characterized in that The real-time access data includes real-time user access behavior data of merchants and real-time user access behavior data of products.

3. The method according to claim 2, characterized in that After the step of sorting the user's historically visited merchants and historically visited commodities according to the historical merchant scores and the historical commodity scores to generate a historical merchant sorting list and a historical commodity sorting list, the method further includes: The historical merchant ranking list and the historical commodity ranking list are updated respectively according to the user's real-time visit merchant behavior data and the user's real-time visit commodity behavior data.

4. The method according to claim 3, characterized in that The user's real-time merchant access behavior data includes a real-time merchant click-through rate, the real-time product access behavior data includes a real-time product click-through rate, and the real-time product access behavior data includes a real-time product click-through rate. The step of updating the historical merchant ranking list and the historical product ranking list respectively according to the user's real-time merchant access behavior data and the user's real-time product access behavior data includes: Update the historical merchant ranking list according to the real-time merchant click-through rate; The historical merchant ranking list is updated according to the real-time commodity click-through rate.

5. The method according to claim 1, wherein The first machine learning model is trained by the following steps: Extracting attribute data of each user, attribute data of each merchant, and visit rate of each user to each merchant within a preset time period; Scoring each merchant according to the visit rate of each user to each merchant, and obtaining visit characteristic data of each user to each merchant; Performing feature conversion on the attribute data of each user, the attribute data of each merchant, and the characteristic data of each user's visit to each merchant to obtain a second training feature vector of a preset dimension; Extracting attribute data of each user, attribute data of each product, and access rate of each user to the product within a preset time period; Scoring each commodity according to the visit rate of each user to each commodity, and obtaining visit characteristic data of each user to each commodity; Performing feature conversion on the user attribute data, the product attribute data, and the user access feature data of each product to obtain a product training feature vector of a preset dimension; The second training feature vector and the product training feature vector are respectively trained using a preset machine learning algorithm to obtain a first machine learning model.

6. The method according to claim 1, characterized in that The second machine learning model is trained by the following steps: Extracting access data of all users to merchants or products within a preset time period, wherein the access data includes attribute data of each user, attribute data of each merchant, attribute data of each product, access rate of each user to the merchant, and access rate of each user to the product; Scoring each merchant based on each user's visit behavior data to the merchant, and obtaining characteristic data of each user's visit to the merchant; Scoring each product based on each user's access behavior data to the product, and obtaining characteristic data of each user's access to the product; Performing feature conversion on the user attribute data, the merchant attribute data, the product attribute data, the characteristic data of each user's visit to the merchant, and the characteristic data of each user's visit to the product to generate a first training feature vector of a preset dimension; The first training feature vector is trained using a preset machine learning algorithm to obtain a second machine learning model.

7. A recommendation device, characterized in that: include: A data acquisition module is used to acquire user access data; the access data includes historical access data and / or real-time access data; A recommended merchant product list generation module is used to input the access data into a pre-trained machine learning model to generate a recommended merchant product list for the user, including: Performing feature conversion on the historical access data to obtain a first feature vector of a preset dimension; the historical access data includes user attribute data, merchant attribute data, user historical visit behavior data of merchants, product attribute data, and user historical visit behavior data of products; Performing feature conversion on the real-time access data to obtain a second feature vector of a preset dimension; Inputting the first feature vector into a pre-trained first machine learning model to obtain the user's historical merchant scores and historical product scores respectively; Inputting the second feature vector into a pre-trained second machine learning model to obtain the user's real-time merchant score and real-time product score; Sort the user's historically visited merchants and historically visited commodities according to the historical merchant scores and the historical commodity scores, and generate a historical merchant sorting list and a historical commodity sorting list; Sort the real-time merchants and real-time commodities visited by the user according to the real-time merchant scores and the real-time commodity scores, and generate a real-time merchant ranking list and a real-time commodity ranking list; According to the historical merchant ranking list and the historical product ranking list, respectively modifying the real-time merchant ranking list and the real-time product ranking list to generate a recommended merchant product list for the user, including: Perform weighted summation on the historical merchant ranking list and product weights, and on the historical product ranking list and merchant weights to obtain a merchant product weight ranking list; According to the merchant product order in the merchant product weight ranking list, the real-time merchant ranking list and the real-time product ranking list are modified to generate a recommended merchant product list for the user; The merchant product list includes recommended merchants and recommended products for the recommended merchants; The recommended merchant product list display module is used to display the merchants and products in the recommended merchant product list to the user accordingly.

8. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the recommendation method according to any one of claims 1 to 6 when executing the program.

9. A readable storage medium, characterized in that When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to implement the recommendation method according to any one of claims 1 to 6.

Images