Commodity recommendation method and device

Abstract

The invention discloses a commodity recommendation method and device. The method comprises the following steps: collecting attribute information and behavior information of a target user and attribute and behavior information of all commodities in a commodity pool; determining a target recommended commodity matched with the target user by using a pre-trained commodity recommendation model; the model training sample comprises attribute and behavior information of a plurality of users and attribute and behavior information of a plurality of commodities; analyzing through a collaborative filtering algorithm, and calculating a first similarity between the users and a second similarity between the commodities; according to the first similarity and the second similarity, determining an interested score of each user for each commodity, user preferences and a recommendation priority of each commodity for each user, and determining a corresponding recommended commodity for each user; and based on the attribute information of the target user, selecting a target commodity recommendation mode to recommend a target recommended commodity to the target user. The technical problems that related commodity recommendation schemes are difficult to meet personalized requirements of users and the recommendation effect is poor are solved.

Description

Technical Field

[0001] The present application relates to the field of big data processing technology, and more specifically, to a product recommendation method and device. Background Art

[0002] With the rapid development of the telecommunications industry and the deepening of digital transformation, telecom operators are faced with the challenge of how to effectively manage a huge pool of products and how to accurately provide users with telecom product recommendations. Traditional recommendation systems are often based on relatively simple user behavior analysis, such as browsing history, purchase history, etc., and use content filtering or collaborative filtering algorithms based on users or items for recommendation. However, such recommendation methods have the following problems: Traditional recommendation algorithms often fail to fully consider the personalized needs and preferences of users, which means that the recommended products may not match the actual needs and interests of users, resulting in poor recommendation results and low user satisfaction. User needs and preferences may change over time, environment and other factors. Traditional recommendation algorithms often find it difficult to capture these uncertainties in a timely manner, resulting in inflexible and inaccurate recommendation results. Existing recommendation systems often find it difficult to explain the reasons for recommendations to users, and users lack understanding and trust in the recommendation results, which limits the application scope and user acceptance of recommendation systems.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a product recommendation method and device to at least solve the technical problem that related product recommendation solutions are difficult to meet the personalized needs of users and the recommendation effect is poor.

[0005] According to one aspect of an embodiment of the present application, a product recommendation method is provided, comprising: obtaining target user attribute information and target user behavior information of a target user, and target product attribute information and target product behavior information of all products in a product pool; using a pre-trained product recommendation model to analyze the target user attribute information, target user behavior information, target product attribute information and target product behavior information to obtain target recommended products corresponding to the target user, wherein the training samples for training the product recommendation model include: user attribute information and user behavior information of multiple users, product attribute information and product behavior information of multiple products, and the sample labels for training the product recommendation model are recommended products corresponding to each user determined in the following manner: using a collaborative filtering algorithm to analyze the user attribute information, user behavior information, product attribute information and product behavior information to obtain a first similarity between multiple users and a second similarity between multiple products, and determining the interest score of each user for each product, the user preference, and the recommendation priority of each product for each user based on the first similarity and the second similarity, and determining the recommended products corresponding to each user based on the interest score, the user preference and the recommendation priority; determining the target product recommendation method corresponding to the target user attribute information, and recommending the target recommended products to the target user based on the target product recommendation method.

[0006] Optionally, the training process of the product recommendation model includes: constructing an initial recommendation model based on deep learning; obtaining user attribute information and user behavior information of multiple users, product attribute information and product behavior information of multiple products, and using the user attribute information and user behavior information of each user, and the product attribute information and product behavior information of multiple products as a group of training samples, and determining the sample labels corresponding to each group of training samples; using multiple groups of training samples and sample labels to iteratively train the initial recommendation model, and using the stochastic gradient descent method to optimize the model parameters, and when the preset training termination conditions are met, using the trained model as the product recommendation model.

[0007] Optionally, determining the sample labels corresponding to each group of training samples includes: constructing a user-product co-occurrence matrix based on user behavior information of multiple users and product behavior information of multiple products, wherein the elements in the user-product co-occurrence matrix are comprehensive scores of each user for each product, and the comprehensive score is a score obtained by weighted calculation of the number of different behaviors of each user for each product and the evaluation score; determining a first similarity between every two users and a first interest score of each user for each product by using a user-based collaborative filtering algorithm based on the user-product co-occurrence matrix; determining a second similarity between every two products and a second interest score of each user for each product by using a product-based collaborative filtering algorithm; and determining, for each user and each product, a user preference coefficient for the product based on the first user score and the first user similarity, wherein the first user score is the user-product similarity. The first user similarity is the average of the first similarities between the user and all other users who have taken actions on the product; the recommendation priority coefficient of the product to the user is determined based on the first user score, the first product score, the first user similarity, and the first product similarity, wherein the first product score is the average of the scores of all users on the product in the user-product co-occurrence matrix, and the first product similarity is the average of the second similarities between the product and other products on which the user has taken actions; the recommendation score of the product to the user is determined based on the preference coefficient, the recommendation priority coefficient, the first interest score, and the second interest score; for each user, multiple products are sorted from large to small based on the recommendation scores of multiple products to the user, and a preset number of products with the top rankings are determined as the recommended products corresponding to the user, and the recommended products are used as sample labels of the training samples corresponding to the user.

[0008] Optionally, the comprehensive score is calculated as: ui =ω1·b1+ω2·b2+ω3·b3+ω4·b4+ω5·s4, where, r ui represents the comprehensive score of user u for product i, ω1, ω2, ω3, ω4, ω5 are preset weight coefficients, b1, b2, b3, b4, s4 represent the number of views, purchases, searches, evaluations and evaluation scores of user u for product i respectively; the calculation formula of the first similarity is: In the formula, sim(u,v) represents the first similarity between user u and user v, A represents the set of products that both user u and user v have taken actions on, and r uk represents the comprehensive rating of user u for product k, r vk represents the comprehensive score of user v on product k; the calculation formula of the first interest score is: Where P ui1represents the first interest rating of user u for product i, N(u) represents the set of N users with the highest first similarity to user u, sim(u,n) represents the first similarity between user u and user n, r ni represents the comprehensive score of user n on product i; the calculation formula of the second similarity is: In the formula, sim(i,j) represents the second similarity between product i and product j, B represents the set of users who have taken actions on both product i and product j, and r ni represents the comprehensive rating of user n on product i, r nj represents the comprehensive score of user n on product j; the calculation formula of the second interest score is: Where P ui2 represents the second interest score of user u for product i, N(i) represents the set of N products with the highest second similarity to product i, sim(i,k) represents the second similarity between product i and product k, r uk represents the comprehensive rating of user u to product k; the calculation formula of preference coefficient is: In the formula, α ui represents the preference coefficient of user u for product i, r ui represents the comprehensive score of user u on product i, C represents the set of users who have taken actions on product i except user u, sim(u,n) represents the first similarity between user u and user n, c1 and c2 are preset constants; the calculation formula of the recommendation priority coefficient is: In the formula, β ui represents the recommendation priority coefficient of product i to user u, C represents the set of users other than user u who have taken actions on product i, sim(u,n) represents the first similarity between user u and user n, D represents the set of other products other than product i that user u has taken actions on, sim(i,k) represents the second similarity between product i and product k, and r ui represents the comprehensive score of user u on product i, E represents the set of users who have taken actions on product i, r ni S represents the comprehensive score of user n for product i, c3 and c4 are preset constants; the calculation formula for the recommended score is: ui =α ui β ui ·θ1·P ui1 ·θ2·P ui2 , where S ui represents the recommendation score of product i to user u, α ui represents the preference coefficient of user u for product i, β ui represents the recommendation priority coefficient of product i to user u, P ui1 represents the first interest rating of user u for product i, Pui2 represents the second interest rating of user u for product i, and θ1 and θ2 are preset weight coefficients.

[0009] Optionally, the target user attribute information includes: the target user's identity, age and gender, and determining the target product recommendation method corresponding to the target user attribute information includes: querying the product recommendation method corresponding to the target user's identity from a product recommendation method list, wherein the product recommendation method corresponding to the identity of different users is stored in the product recommendation method list; if there is a product recommendation method corresponding to the target user's identity, using the product recommendation method as the target product recommendation method; if there is no product recommendation method corresponding to the target user's identity, determining a template recommendation method that matches the target user's age and gender from the product recommendation method list as the target product recommendation method; in response to the target user adjusting the template recommendation method, storing the adjusted template recommendation method as the product recommendation method corresponding to the target user's identity in the product recommendation method list.

[0010] Optionally, after recommending the target recommended product to the target user based on the target product recommendation method, the method also includes: obtaining the target user's click-through rate and purchase conversion rate for the target recommended product, and determining the ratio of the purchase conversion rate to the click-through rate; using the target user attribute information, the target user behavior information, the target product attribute information and the target product behavior information as a new training sample, and in the process of determining the sample label of the new training sample, when determining the recommendation score of the target recommended product for the target user, using the ratio as an additional adjustment coefficient to determine the recommendation score of the target recommended product for the target user; periodically using the new training samples and the corresponding sample labels to iteratively train the product recommendation model and update the model parameters.

[0011] Optionally, user attribute information includes at least one of the following: the user's age, gender, address, and completed services; user behavior information includes at least one of the following: the user's browsing history, purchase history, search history, and evaluation history; product attribute information includes at least one of the following: the product's type, price, brand, and network standard; product behavior information includes at least one of the following: the product's browsing history, purchase history, search history, and evaluation history.

[0012] According to another aspect of an embodiment of the present application, a product recommendation device is also provided, including: an acquisition module, used to acquire target user attribute information and target user behavior information of a target user, and target product attribute information and target product behavior information of all products in a product pool; an analysis module, used to analyze the target user attribute information, target user behavior information, target product attribute information and target product behavior information using a pre-trained product recommendation model to obtain a target recommended product corresponding to the target user, wherein the training samples for training the product recommendation model include: user attribute information and user behavior information of multiple users, product attribute information and product behavior information of multiple products, and the training product recommendation model The sample labels are the recommended products corresponding to each user determined in the following manner: using a collaborative filtering algorithm to analyze user attribute information, user behavior information, product attribute information and product behavior information, obtaining a first similarity between multiple users and a second similarity between multiple products, and determining each user's interest score for each product, user preference, and recommendation priority of each product to each user based on the first similarity and the second similarity, and determining the recommended products corresponding to each user based on the interest score, user preference and recommendation priority; a recommendation module is used to determine a target product recommendation method corresponding to the target user attribute information, and recommend the target recommended product to the target user based on the target product recommendation method.

[0013] According to another aspect of an embodiment of the present application, a computer program product is further provided, the computer program product comprising: a computer program, wherein when the computer program is executed by a processor, the above-mentioned commodity recommendation method is implemented.

[0014] According to another aspect of an embodiment of the present application, an electronic device is further provided, comprising: a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the above-mentioned product recommendation method through the computer program.

[0015] In an embodiment of the present application, by collecting target user attribute information (including age, gender, identity, etc.) and behavior information (such as purchase, browsing history, etc.), combined with the detailed attributes and market behavior data of the products in the product pool, a pre-trained product recommendation model can generate highly personalized product recommendations. For the pre-trained product recommendation model, during the training process of the model, a collaborative filtering algorithm is used to calculate the first similarity between users and the second similarity between products, which not only helps the model identify common preferences in user groups, but also understands the correlation between products. By combining similarity with user product interest ratings, user preferences and product recommendation priorities, the model can generate more optimized recommendation decisions to ensure that the recommended products are It not only meets the personalized needs of users, but also has high market appeal and potential value. The model design includes a dynamic feedback mechanism, which can learn and adjust from the actual feedback of target users on recommended products (such as click-through rate and purchase conversion rate). By periodically using these feedback information to iterate model training, it can continuously update model parameters and improve prediction accuracy and recommendation effects. The model not only optimizes product recommendations, but also supports the determination of personalized recommendation methods. It can intelligently match the most suitable recommendation channel based on user identity, age, gender and other information, making the recommendation service more in line with user habits and preferences, thereby solving the technical problem that related product recommendation schemes are difficult to meet user personalized needs and the recommendation effect is poor. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 is a flowchart of an optional commodity recommendation method according to an embodiment of the present application;

[0018] Figure 2 is a schematic structural diagram of an optional commodity recommendation device according to an embodiment of the present application;

[0019] Figure 3 It is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] 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.

[0021] It should be noted that the terms "first", "second", etc. in the specification, claims and drawings of the present application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. 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.

[0022] In order to better understand the embodiments of the present application, some nouns or terms that appear in the description of the embodiments of the present application are first translated and explained as follows:

[0023] Collaborative Filtering (CF) algorithm: A technology widely used in recommendation systems, which aims to provide users with personalized recommendations by analyzing their historical behaviors and preferences. The core idea of ​​collaborative filtering is that "similar users like similar items" or "similar items are liked by similar users". Depending on the implementation method, collaborative filtering can be divided into user-based collaborative filtering (User-based CF) and item-based collaborative filtering (Item-based CF), as well as more complex variants such as matrix decomposition.

[0024] Example 1

[0025] According to an embodiment of the present application, a product recommendation method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0026] Figure 1 is a flowchart of a product recommendation method provided according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0027] Step S102, obtaining target user attribute information and target user behavior information of the target user, and target commodity attribute information and target commodity behavior information of all commodities in the commodity pool;

[0028] Step S104, using a pre-trained product recommendation model to analyze target user attribute information, target user behavior information, target product attribute information and target product behavior information to obtain target recommended products corresponding to the target user, wherein the training samples for training the product recommendation model include: user attribute information and user behavior information of multiple users, product attribute information and product behavior information of multiple products, and the sample labels for training the product recommendation model are recommended products corresponding to each user determined by the following method: using a collaborative filtering algorithm to analyze user attribute information, user behavior information, product attribute information and product behavior information to obtain a first similarity between multiple users and a second similarity between multiple products, and determining the interest score of each user for each product, the user preference, and the recommendation priority of each product for each user based on the first similarity and the second similarity, and determining the recommended product corresponding to each user based on the interest score, user preference and recommendation priority;

[0029] Step S106: determine a target commodity recommendation method corresponding to the target user attribute information, and recommend a target recommended commodity to the target user based on the target commodity recommendation method.

[0030] The following describes each step of the product recommendation method in conjunction with a specific implementation process.

[0031] First, the target user attribute information and target user behavior information of the target user, and the target product attribute information and target product behavior information of all products in the product pool are obtained, so as to generate more accurate product recommendations by deeply understanding the characteristics of the target user and all products in the product pool.

[0032] As an optional implementation, user attribute information includes at least one of the following: the user's age, gender, address, and completed services; user behavior information includes at least one of the following: the user's browsing history, purchase history, search history, and evaluation history; product attribute information includes at least one of the following: the product's type, price, brand, and network standard; product behavior information includes at least one of the following: the product's browsing history, purchase history, search history, and evaluation history.

[0033] After obtaining the above information, the target user attribute information, target user behavior information, target product attribute information and target product behavior information are analyzed using a pre-trained product recommendation model to obtain target recommended products corresponding to the target user, wherein the training samples for training the product recommendation model include: user attribute information and user behavior information of multiple users, product attribute information and product behavior information of multiple products, and the sample labels for training the product recommendation model are recommended products corresponding to each user determined in the following manner: using a collaborative filtering algorithm to analyze user attribute information, user behavior information, product attribute information and product behavior information to obtain a first similarity between multiple users and a second similarity between multiple products, and determining each user's interest score for each product, user preference, and recommendation priority of each product for each user based on the first similarity and the second similarity, and determining the recommended product corresponding to each user based on the interest score, user preference and recommendation priority.

[0034] In general, the user-based collaborative filtering method has the following conventional steps: Calculate the similarity between the target user and other users, and commonly used methods include cosine similarity, Pearson correlation coefficient, etc.; Based on the ratings of similar users, predict the current user's rating for the rated items, usually using the weighted average method, with the weight being the similarity between users; Based on the predicted ratings, recommend the highest-rated items to the user. The commodity-based collaborative filtering algorithm has the following conventional steps: Calculate the similarity between items, and commonly used methods are also cosine similarity, Pearson correlation coefficient, etc.; Based on the user's ratings of similar items, predict the user's rating for unrated items, and also use the weighted average method, with the weight being the similarity between items; Based on the predicted ratings, recommend the highest-rated items to the user. This embodiment uses the corresponding collaborative filtering algorithm when calculating the first similarity and the second similarity.

[0035] As an optional implementation, the training process of the product recommendation model can be carried out in the following steps:

[0036] S1, build an initial recommendation model based on deep learning;

[0037] S2, obtaining user attribute information and user behavior information of multiple users, and commodity attribute information and commodity behavior information of multiple commodities, and using the user attribute information and user behavior information of each user, and the commodity attribute information and commodity behavior information of multiple commodities as a group of training samples, and determining a sample label corresponding to each group of training samples;

[0038] S3, iteratively trains the initial recommendation model using multiple sets of training samples and sample labels, and optimizes the model parameters using the stochastic gradient descent method. When the preset training termination conditions are met, such as the number of iterations reaches a threshold or the gradient no longer decreases, the trained model is used as the product recommendation model.

[0039] As an optional implementation, during model training, the following process may be used to determine the sample labels corresponding to each group of training samples:

[0040] S1, construct a user-product co-occurrence matrix based on the user behavior information of multiple users and the product behavior information of multiple products. The elements in the user-product co-occurrence matrix are the comprehensive scores of each user for each product. The comprehensive score is the score obtained by weighted calculation of the number of different behaviors of each user for each product and the evaluation score. The different behaviors of the user for each product refer to the user's browsing, purchasing, searching or evaluating behaviors for each product.

[0041] The user-product co-occurrence matrix is ​​a data structure used to reflect the strength of the association between users and products. The rows of the matrix represent users, the columns represent products, and the element values ​​represent the user's comprehensive rating of the product. The comprehensive rating is obtained by weighted calculation of the user's different behaviors on the product (such as the number of purchases, the number of views, etc.) and the evaluation score (such as a five-star rating). For example, a user may have browsed product A many times but did not buy it, giving it a high rating, purchased product B once, gave it a low rating, and did not take any action on product C. This information will be integrated into the user-product co-occurrence matrix for subsequent analysis. As an optional implementation, the calculation formula for the comprehensive rating is:

[0042] r ui =ω1·b1+ω2·b2+ω3·b3+ω4·b4+ω5·s4

[0043] In the formula, r ui represents the comprehensive score of user u for product i, ω1, ω2, ω3, ω4, and ω5 are preset weight coefficients respectively, and b1, b2, b3, b4, and s4 represent the number of views, purchases, searches, evaluations, and evaluation scores of user u for product i respectively.

[0044] S2, according to the user-product co-occurrence matrix, a user-based collaborative filtering algorithm is used to determine the first similarity between every two users and the first interest score of each user for each product. The similarity is based on the similarity of user behavior. For example, if two users often buy the same product, the similarity between them is high. The interest score is based on the user's direct score of the product and the frequency of behavior. In the embodiment of the present application, the calculation formula of the first similarity is:

[0045]

[0046] In the formula, sim(u,v) represents the first similarity between user u and user v, A represents the set of products that both user u and user v have taken actions on, and r ukrepresents the comprehensive rating of user u for product k, r vk Represents the comprehensive rating of user v to product k.

[0047] The calculation formula for the first interest score is:

[0048]

[0049] Where P ui1 represents the first interest rating of user u for product i, N(u) represents the set of N users with the highest first similarity to user u, sim(u,n) represents the first similarity between user u and user n, r ni Represents the comprehensive rating of user n on product i.

[0050] S3, using a collaborative filtering algorithm based on commodities to determine a second similarity between every two commodities and a second interest score of each user for each commodity.

[0051] The calculation formula of the second similarity is:

[0052]

[0053] In the formula, sim(i,j) represents the second similarity between product i and product j, B represents the set of users who have taken actions on both product i and product j, and r ni represents the comprehensive rating of user n on product i, r nj represents the comprehensive rating of user n on product j.

[0054] The calculation formula for the second interest score is:

[0055]

[0056] Where P ui2 represents the second interest score of user u for product i, N(i) represents the set of N products with the highest second similarity to product i, sim(i,k) represents the second similarity between product i and product k, r uk Represents the comprehensive rating of user u for product k.

[0057] S4, for each user and each product, determine the user's preference coefficient for the product based on the first user score and the first user similarity, wherein the first user score is the user's comprehensive score for the product in the user-product co-occurrence matrix, and the first user similarity is the average of the first similarities between the user and all other users who have behaviors on the product. The calculation formula of the preference coefficient in the embodiment of the present application is:

[0058]

[0059] In the formula, α ui represents the preference coefficient of user u for product i, r ui represents the comprehensive score of user u on product i, C represents the set of users other than user u who have taken actions on product i, sim(u,n) represents the first similarity between user u and user n, and c1 and c2 are preset constants.

[0060] Among them, the molecule r ui It represents the comprehensive score of user u for product i. This score directly reflects the user's preference for the product. The higher the score, the greater the user's interest in the product. c1 is added to the denominator to prevent the denominator from being 0 (that is, when user u is completely dissimilar to other users, avoid dividing by 0). In addition, c1 can also adjust the sensitivity of the public display, so that even if the similarity is low, it will not lead to a too large preference coefficient. Overall, The meaning of is to divide the user's comprehensive score by the average similarity between user u and other users who have taken actions on product i. If user u is very similar to other users who have taken actions on product i (that is, the average similarity is high), then the score of user u may be affected by these similar users, and thus the preference coefficient will be relatively small. On the contrary, if user u is not similar to other users, then the score of user u can better represent personal preference, and the preference coefficient will be larger. The existence of c2 ensures that the preference coefficient will not be lower than the preset lower limit, which means that even if the user's score is low or the similarity with other users is high, the preference coefficient will not be too small, thereby ensuring a certain minimum preference value, which helps to avoid unreasonable recommendations in extreme cases.

[0061] S5, determining the recommendation priority coefficient of the product to the user according to the first user score, the first product score, the first user similarity, and the first product similarity, wherein the first product score is the average of the scores of all users on the product in the user-product co-occurrence matrix, the first product similarity is the average of the second similarities between the product and other products that the user has behavior, and the calculation formula of the recommendation priority coefficient is:

[0062]

[0063] In the formula, β ui represents the recommendation priority coefficient of product i to user u, C represents the set of users other than user u who have taken actions on product i, sim(u,n) represents the first similarity between user u and user n, D represents the set of other products other than product i that user u has taken actions on, sim(i,k) represents the second similarity between product i and product k, and r ui represents the comprehensive score of user u on product i, E represents the set of users who have taken actions on product i, r niIt represents the comprehensive rating of user n to product i, and c3 and c4 are preset constants.

[0064] It should be noted that It represents the average similarity between user u and other users who have taken actions on product i. It reflects the overall similarity between user u and other users who have taken actions on product i. It represents the average similarity between product i and other products that user u has interacted with. It reflects the similarity between product i and other products that user u has interacted with, and helps capture the user's interest in similar products. It represents the average rating of all users who have taken actions on product i. It reflects the overall popularity of product i, that is, the average evaluation of product i by all users; represents the product of user similarity and product similarity. This product term emphasizes the combined effect of the dual similarity between users and products. That is, if user u is very similar to other users who have taken actions on product i, and product i is also very similar to other products that user u has previously come into contact with, then product i will have a higher recommendation priority for user u. It represents the product of the user's rating of product i and the average rating of product i. This product term takes into account the user's personal preference for product i and the overall popularity of product i. If the user has a high rating for product i and product i is also generally well received by other users, then the recommendation priority of the product will be higher. c3 is mainly to prevent the numerator from being zero, and can adjust the sensitivity of the recommendation priority coefficient.

[0065] S6, determining the recommendation score of the product for the user based on the preference coefficient, the recommendation priority coefficient, the first interest score and the second interest score. The calculation formula of the recommendation score is:

[0066] S ui =α ui β ui ·θ1·P ui1 ·θ2·P ui2

[0067] In the formula, S ui represents the recommendation score of product i to user u, α ui represents the preference coefficient of user u for product i, β ui represents the recommendation priority coefficient of product i to user u, P ui1 represents the first interest rating of user u for product i, P ui2 represents the second interest rating of user u for product i, and θ1 and θ2 are preset weight coefficients.

[0068] S7, for each user, sort the multiple commodities from large to small according to the recommendation scores of the multiple commodities to the user, determine a preset number of commodities with the highest ranking as recommended commodities corresponding to the user, and use the recommended commodities as sample labels of the training samples corresponding to the user.

[0069] After the product recommendation model is trained in the above manner, the model is used to analyze the target user attribute information, target user behavior information, target product attribute information and target product behavior information to obtain the target recommended product corresponding to the target user.

[0070] After obtaining the target recommended product corresponding to the target user, a target product recommendation method corresponding to the target user attribute information is determined, and the target recommended product is recommended to the target user based on the target product recommendation method.

[0071] As an optional implementation, the target user attribute information includes: the target user's identity, age and gender. The process of determining the target product recommendation method corresponding to the target user's attribute information can take the following steps: querying the product recommendation method corresponding to the target user's identity from a product recommendation method list, wherein the product recommendation methods corresponding to the identity identifiers of different users are stored in the product recommendation method list; if there is a product recommendation method corresponding to the target user's identity identifier, the product recommendation method is used as the target product recommendation method; if there is no product recommendation method corresponding to the target user's identity identifier, a template recommendation method matching the target user's age and gender is determined from the product recommendation method list as the target product recommendation method, wherein the template recommendation method is a recommendation strategy pre-set based on the common preferences of the user group. For example, young users may prefer to receive recommendations through mobile applications, while older users may prefer emails or text messages; in response to the target user's adjustment of the template recommendation method, the adjusted template recommendation method is stored in the product recommendation method list as the product recommendation method corresponding to the target user's identity identifier.

[0072] As an optional implementation, after recommending the target recommended product to the target user based on the target product recommendation method, the following steps may be performed to further collect the target user's feedback data on the recommended product to optimize the performance of the recommendation model:

[0073] Obtain the target user's click-through rate and purchase conversion rate for the target recommended product, and determine the ratio of the purchase conversion rate to the click-through rate; use the target user attribute information, target user behavior information, target product attribute information, and target product behavior information as a new training sample, and in the process of determining the sample label of the new training sample, when determining the recommendation score of the target recommended product for the target user, use the ratio as an additional adjustment coefficient to determine the recommendation score of the target recommended product for the target user, periodically use the new training samples and the corresponding sample labels to iteratively train the product recommendation model and update the model parameters.

[0074] Among them, the click-through rate refers to the ratio of the number of times a user clicks on a product in the recommendation list to the total number of times the product is displayed to the user. In the scenario of telecommunications product recommendation, the click-through rate reflects the user's interest in the recommended product and is a quantitative indicator of the initial response to the recommendation result. The purchase conversion rate refers to the proportion of users who complete the purchase behavior after clicking on the product, that is, the ratio of the number of purchases to the number of clicks. For telecommunications product recommendation, the purchase conversion rate is a key indicator to measure whether the recommended product really meets the user's needs and whether it can effectively promote sales. For example, if a product has a high click-through rate but a relatively low purchase conversion rate, it may indicate that although the user is interested in the product, the possibility of ultimately deciding to buy it is low. On the contrary, if the ratio of the purchase conversion rate to the click-through rate of a product is high, it means that once the user clicks, the probability of purchase is high, and such a product is more attractive to the user. The ratio of the purchase conversion rate to the click-through rate is used as an additional adjustment coefficient. This coefficient is used in the calculation of the subsequent recommendation score, which will reflect the actual feedback effect of user behavior, that is, the real response degree of the user to the recommended product.

[0075] In the optimization process of the recommendation model, the ratio of purchase conversion rate to click-through rate is used as an additional adjustment coefficient to adjust the recommendation score and evaluate the quality of the recommendation results. A higher ratio means that users have a stronger actual purchase intention for the recommended products, reflecting the effectiveness of the product recommendations. By incorporating this ratio into model training, the model parameters can be adjusted to pay more attention to products with high conversion rates, thereby optimizing the recommendation strategy and ensuring that the recommendation algorithm can adapt to changes in user behavior patterns in a timely manner, improve the personalization and timeliness of recommendations, as well as the overall recommendation effect and user conversion rate.

[0076] In an embodiment of the present application, by collecting target user attribute information (including age, gender, identity, etc.) and behavior information (such as purchase, browsing history, etc.), combined with the detailed attributes and market behavior data of the products in the product pool, a pre-trained product recommendation model can generate highly personalized product recommendations. For the pre-trained product recommendation model, during the training process of the model, a collaborative filtering algorithm is used to calculate the first similarity between users and the second similarity between products, which not only helps the model identify common preferences in user groups, but also understands the correlation between products. By combining similarity with user product interest ratings, user preferences and product recommendation priorities, the model can generate more optimized recommendation decisions to ensure that the recommended products are It not only meets the personalized needs of users, but also has high market appeal and potential value. The model design includes a dynamic feedback mechanism, which can learn and adjust from the actual feedback of target users on recommended products (such as click-through rate and purchase conversion rate). By periodically using these feedback information to iterate model training, it can continuously update model parameters and improve prediction accuracy and recommendation effects. The model not only optimizes product recommendations, but also supports the determination of personalized recommendation methods. It can intelligently match the most suitable recommendation channel based on user identity, age, gender and other information, making the recommendation service more in line with user habits and preferences, thereby solving the technical problem that related product recommendation schemes are difficult to meet user personalized needs and the recommendation effect is poor.

[0077] Example 2

[0078] According to an embodiment of the present application, a commodity recommendation device for implementing the commodity recommendation method in Embodiment 1 is also provided. Figure 2 As shown, the product recommendation device at least includes: an acquisition module 21, an analysis module 22 and a recommendation module 23, wherein:

[0079] An acquisition module 21 is used to acquire target user attribute information and target user behavior information of a target user, and target commodity attribute information and target commodity behavior information of all commodities in a commodity pool;

[0080] The analysis module 22 is used to analyze the target user attribute information, the target user behavior information, the target product attribute information and the target product behavior information by using the pre-trained product recommendation model to obtain the target recommended product corresponding to the target user, wherein the training samples for training the product recommendation model include: user attribute information and user behavior information of multiple users, product attribute information and product behavior information of multiple products, and the sample labels for training the product recommendation model are the recommended products corresponding to each user determined by the following method: using the collaborative filtering algorithm to analyze the user attribute information, user behavior information, product attribute information and product behavior information to obtain the first similarity between multiple users and the second similarity between multiple products, and determining the interest score of each user for each product, the user preference, and the recommendation priority of each product for each user based on the first similarity and the second similarity, and determining the recommended product corresponding to each user based on the interest score, user preference and recommendation priority;

[0081] The recommendation module 23 is used to determine a target commodity recommendation method corresponding to the target user attribute information, and recommend a target recommended commodity to the target user based on the target commodity recommendation method.

[0082] The functions of each module of the product recommendation device are described below in conjunction with a specific implementation process.

[0083] The acquisition module acquires the target user attribute information and target user behavior information of the target user, and the target product attribute information and target product behavior information of all products in the product pool, thereby generating more accurate product recommendations by deeply understanding the characteristics of the target user and all products in the product pool;

[0084] As an optional implementation, user attribute information includes at least one of the following: the user's age, gender, address, and completed services; user behavior information includes at least one of the following: the user's browsing history, purchase history, search history, and evaluation history; product attribute information includes at least one of the following: the product's type, price, brand, and network standard; product behavior information includes at least one of the following: the product's browsing history, purchase history, search history, and evaluation history.

[0085] After obtaining the above information, the analysis module uses the pre-trained product recommendation model to analyze the target user attribute information, target user behavior information, target product attribute information and target product behavior information to obtain the target recommended product corresponding to the target user, wherein the training samples for training the product recommendation model include: user attribute information and user behavior information of multiple users, product attribute information and product behavior information of multiple products, and the sample labels for training the product recommendation model are the recommended products corresponding to each user determined by the following method: using a collaborative filtering algorithm to analyze the user attribute information, user behavior information, product attribute information and product behavior information to obtain a first similarity between multiple users and a second similarity between multiple products, and determining the interest score of each user for each product, the user preference, and the recommendation priority of each product for each user based on the first similarity and the second similarity, and determining the recommended product corresponding to each user based on the interest score, user preference and recommendation priority;

[0086] As an optional implementation, the training process of the product recommendation model can be carried out in the following steps:

[0087] S1, build an initial recommendation model based on deep learning;

[0088] S2, obtaining user attribute information and user behavior information of multiple users, and commodity attribute information and commodity behavior information of multiple commodities, and using the user attribute information and user behavior information of each user, and the commodity attribute information and commodity behavior information of multiple commodities as a group of training samples, and determining a sample label corresponding to each group of training samples;

[0089] S3, iteratively trains the initial recommendation model using multiple sets of training samples and sample labels, and optimizes the model parameters using the stochastic gradient descent method. When the preset training termination conditions are met, the trained model is used as the product recommendation model.

[0090] As an optional implementation, during model training, the following process may be used to determine the sample labels corresponding to each group of training samples:

[0091] S1, construct a user-product co-occurrence matrix based on the user behavior information of multiple users and the product behavior information of multiple products. The elements in the user-product co-occurrence matrix are the comprehensive scores of each user for each product. The comprehensive score is the score obtained by weighted calculation of the number of different behaviors of each user for each product and the evaluation score.

[0092] As an optional implementation, the calculation formula of the comprehensive score is:

[0093] r ui =ω1·b1+ω2·b2+ω3·b3+ω4·b4+ω5·s4

[0094] In the formula, r ui represents the comprehensive score of user u for product i, ω1, ω2, ω3, ω4, ω5 are preset weight coefficients, b1, b2, b3, b4, s4 represent the number of views, purchases, searches, evaluations and evaluation scores of user u for product i respectively;

[0095] S2, according to the user-product co-occurrence matrix, using a user-based collaborative filtering algorithm to determine a first similarity between every two users and a first interest score of each user for each product.

[0096] The calculation formula for the first similarity is:

[0097]

[0098] In the formula, sim(u,v) represents the first similarity between user u and user v, A represents the set of products that both user u and user v have taken actions on, and r uk represents the comprehensive rating of user u for product k, r vk represents the comprehensive rating of user v to product k, ∑ k∈A r uk ·r vk It represents the dot product of the ratings of user u and user v on all items k. The larger the dot product, the more similar the ratings of the two users on these items. The Euclidean norm (i.e., the length of the vector) representing the rating vectors of user u and user v is divided by these two norms to normalize the dot product to a value between -1 and 1. This is the cosine similarity. The closer the cosine similarity is to 1, the more similar the behaviors of the two users are.

[0099] The calculation formula for the first interest score is:

[0100]

[0101] Where P ui1 represents the first interest rating of user u for product i. This rating is predicted based on the ratings of other users similar to user u for product i. That is, by finding other users with similar behaviors to user u and inferring the possible interest of user u in product i based on their ratings, N(u) represents the set of N users with the highest first similarity to user u, sim(u,n) represents the first similarity between user u and user n, r ni Represents the comprehensive rating of user n on product i.

[0102] S3, using a collaborative filtering algorithm based on commodities to determine the second similarity between every two commodities and the second interest score of each user for each commodity;

[0103] The calculation formula for the second similarity is:

[0104]

[0105] In the formula, sim(i,j) represents the second similarity between product i and product j, B represents the set of users who have taken actions on both product i and product j, and r ni represents the comprehensive rating of user n on product i, r nj represents the comprehensive rating of user n on product j.

[0106] The calculation formula for the second interest score is:

[0107]

[0108] Where P ui2 represents the second interest score of user u for product i, N(i) represents the set of N products with the highest second similarity to product i, sim(i,k) represents the second similarity between product i and product k, r uk represents the comprehensive rating of user u for product k.

[0109] S4, for each user and each product, determine the user's preference coefficient for the product based on the first user score and the first user similarity, wherein the first user score is the comprehensive score of the user for the product in the user-product co-occurrence matrix, and the first user similarity is the average of the first similarities between the user and all other users who have taken actions on the product.

[0110] The calculation formula of preference coefficient is:

[0111]

[0112] In the formula, α ui represents the preference coefficient of user u for product i, r ui represents the comprehensive score of user u on product i, C represents the set of users other than user u who have taken actions on product i, sim(u,n) represents the first similarity between user u and user n, and c1 and c2 are preset constants;

[0113] S5. Determine a recommendation priority coefficient of the product for the user based on the first user score, the first product score, the first user similarity, and the first product similarity, wherein the first product score is the average of the scores of all users on the product in the user-product co-occurrence matrix, and the first product similarity is the average of the second similarities between the product and other products that the user has generated behavior.

[0114] The formula for calculating the recommended priority coefficient is:

[0115]

[0116] In the formula, β ui represents the recommendation priority coefficient of product i to user u, C represents the set of users other than user u who have taken actions on product i, sim(u,n) represents the first similarity between user u and user n, D represents the set of other products other than product i that user u has taken actions on, sim(i,k) represents the second similarity between product i and product k, and r ui represents the comprehensive score of user u on product i, E represents the set of users who have taken actions on product i, r ni represents the comprehensive rating of product i by user n, c3 and c4 are preset constants;

[0117] S6, determining a recommendation score of the product for the user according to the preference coefficient, the recommendation priority coefficient, the first interest score, and the second interest score.

[0118] The formula for calculating the recommendation score is:

[0119] S ui =α ui β ui ·θ1·P ui1 ·θ2·P ui2

[0120] In the formula, S ui represents the recommendation score of product i to user u, α ui represents the preference coefficient of user u for product i, β ui represents the recommendation priority coefficient of product i to user u, P ui1 represents the first interest rating of user u for product i, P ui2 represents the second interest rating of user u for product i, and θ1 and θ2 are preset weight coefficients.

[0121] S7, for each user, sort the multiple commodities from large to small according to the recommendation scores of the multiple commodities to the user, determine a preset number of commodities with the highest ranking as recommended commodities corresponding to the user, and use the recommended commodities as sample labels of the training samples corresponding to the user.

[0122] After the product recommendation model is trained in the above manner, the model is used to analyze the target user attribute information, target user behavior information, target product attribute information and target product behavior information to obtain the target recommended product corresponding to the target user.

[0123] After obtaining the target recommended product corresponding to the target user, the recommendation module determines a target product recommendation method corresponding to the target user attribute information, and recommends the target recommended product to the target user based on the target product recommendation method.

[0124] Among them, the target user attribute information includes: the target user's identity, age and gender. The process of determining the target product recommendation method corresponding to the target user attribute information can take the following steps: query the product recommendation method corresponding to the target user's identity from the product recommendation method list, wherein the product recommendation method corresponding to the identity of different users is stored in the product recommendation method list; if there is a product recommendation method corresponding to the target user's identity, the product recommendation method is used as the target product recommendation method; if there is no product recommendation method corresponding to the target user's identity, a template recommendation method matching the target user's age and gender is determined from the product recommendation method list as the target product recommendation method; in response to the target user adjusting the template recommendation method, the adjusted template recommendation method is stored in the product recommendation method list as the product recommendation method corresponding to the target user's identity.

[0125] As an optional implementation, after recommending the target recommended product to the target user based on the target product recommendation method, the following steps may be performed to further collect the target user's feedback data on the recommended product to optimize the performance of the recommendation model:

[0126] Obtain the target user's click-through rate and purchase conversion rate for the target recommended product, and determine the ratio of the purchase conversion rate to the click-through rate; use the target user attribute information, target user behavior information, target product attribute information and target product behavior information as a new training sample, and in the process of determining the sample label of the new training sample, when determining the recommendation score of the target recommended product for the target user, use the ratio as an additional adjustment coefficient to determine the recommendation score of the target recommended product for the target user; periodically use the new training samples and the corresponding sample labels to iteratively train the product recommendation model and update the model parameters.

[0127] By introducing actual feedback from user behavior (click-through rate and purchase conversion rate), the system can periodically adjust and optimize the product recommendation model. This mechanism ensures that the recommendation algorithm can adapt to changes in user behavior patterns in a timely manner, improving the personalization and timeliness of recommendations. At the same time, by using user feedback as an additional adjustment factor, the model can more accurately predict user interest in products, thereby improving the conversion rate of recommended products and user satisfaction.

[0128] It should be noted that each module in the product recommendation device in the embodiment of the present application corresponds one by one to each implementation step of the product recommendation method in Example 1. Since a detailed description has been given in Example 1, some details not reflected in this embodiment can be referred to Example 1 and will not be repeated here.

[0129] Example 3

[0130] According to an embodiment of the present application, a computer program product is also provided, which includes a computer program, wherein when the computer program is executed by a processor, the product recommendation method in Example 1 is implemented.

[0131] According to an embodiment of the present application, a non-volatile storage medium is also provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the product recommendation method in Example 1 by running the computer program.

[0132] According to an embodiment of the present application, a processor is also provided, which is used to run a computer program, wherein the product recommendation method in Example 1 is executed when the computer program is running.

[0133] According to an embodiment of the present application, an electronic device is also provided, which includes: a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the product recommendation method in Example 1 through the computer program.

[0134] Specifically, the computer program executes the following steps when it is running: obtaining target user attribute information and target user behavior information of the target user, and target product attribute information and target product behavior information of all products in the product pool; using a pre-trained product recommendation model to analyze the target user attribute information, target user behavior information, target product attribute information and target product behavior information to obtain target recommended products corresponding to the target user, wherein the training samples for training the product recommendation model include: user attribute information and user behavior information of multiple users, product attribute information and product behavior information of multiple products, and the sample labels for training the product recommendation model are recommended products corresponding to each user determined in the following manner: using a collaborative filtering algorithm to analyze the user attribute information, user behavior information, product attribute information and product behavior information to obtain a first similarity between multiple users and a second similarity between multiple products, and determining the interest score of each user for each product, the user preference, and the recommendation priority of each product to each user based on the first similarity and the second similarity, and determining the recommended products corresponding to each user based on the interest score, user preference and recommendation priority; determining the target product recommendation method corresponding to the target user attribute information, and recommending the target recommended products to the target user based on the target product recommendation method.

[0135] As an optional implementation, the electronic device may be in the form of a mobile terminal, a computer terminal or a similar computing device. Figure 3 FIG. 1 shows a hardware structure block diagram of an electronic device for implementing a commodity recommendation method. Figure 3 As shown, the electronic device 30 may include one or more (302a, 302b, ..., 302n are used to illustrate) processors 302 (the processor 302 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 304 for storing data, and a transmission device 306 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 3 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 3 More or fewer components as shown, or with Figure 3 Different configurations shown.

[0136] It should be noted that the one or more processors 302 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the electronic device 30. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0137] The memory 304 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the commodity recommendation method in the embodiment of the present application. The processor 302 executes various functional applications and data processing by running the software programs and modules stored in the memory 304, that is, realizing the vulnerability detection method of the above-mentioned application. The memory 304 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 304 may further include a memory remotely arranged relative to the processor 302, and these remote memories may be connected to the electronic device 30 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0138] The transmission device 306 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the electronic device 30. In one example, the transmission device 306 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 306 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0139] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the electronic device 30 .

[0140] The serial numbers of the above embodiments are only for description and do not represent the advantages or disadvantages of the embodiments.

[0141] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0142] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0143] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0144] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0145] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0146] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A product recommendation method, characterized in that: include: Obtain target user attribute information and target user behavior information of the target user, and target commodity attribute information and target commodity behavior information of all commodities in the commodity pool; Analyze the target user attribute information, the target user behavior information, the target product attribute information and the target product behavior information using a pre-trained product recommendation model to obtain target recommended products corresponding to the target user, wherein the training samples for training the product recommendation model include: user attribute information and user behavior information of multiple users, product attribute information and product behavior information of multiple products, and the sample labels for training the product recommendation model are recommended products corresponding to each user determined in the following manner: analyze the user attribute information, the user behavior information, the product attribute information and the product behavior information using a collaborative filtering algorithm to obtain a first similarity between the multiple users and a second similarity between the multiple products, and determine the interest score of each user for each product, the user preference, and the recommendation priority of each product for each user based on the first similarity and the second similarity, and determine the recommended product corresponding to each user based on the interest score, the user preference and the recommendation priority; A target commodity recommendation method corresponding to the target user attribute information is determined, and the target recommended commodity is recommended to the target user based on the target commodity recommendation method.

2. The method according to claim 1, characterized in that The training process of the product recommendation model includes: Build an initial recommendation model based on deep learning; Acquire user attribute information and user behavior information of multiple users, and commodity attribute information and commodity behavior information of multiple commodities, and use the user attribute information and user behavior information of each user, and the commodity attribute information and commodity behavior information of the multiple commodities as a group of training samples, and determine a sample label corresponding to each group of the training samples; The initial recommendation model is iteratively trained using multiple groups of the training samples and the sample labels, and the model parameters are optimized using a stochastic gradient descent method. When a preset training termination condition is met, the trained model is used as the product recommendation model.

3. The method according to claim 2, characterized in that Determining the sample labels corresponding to each group of the training samples includes: Constructing a user-product co-occurrence matrix based on the user behavior information of the multiple users and the product behavior information of the multiple products, wherein the elements in the user-product co-occurrence matrix are comprehensive scores of each user for each product, and the comprehensive score is a score obtained by weighted calculation of the number of different behaviors of each user for each product and the evaluation score; According to the user-item co-occurrence matrix, a user-based collaborative filtering algorithm is used to determine a first similarity between every two users and a first interest score of each user for each item; a product-based collaborative filtering algorithm is used to determine a second similarity between every two items and a second interest score of each user for each item; For each user and each product, determine the user's preference coefficient for the product based on the first user score and the first user similarity, wherein the first user score is the comprehensive score of the user for the product in the user-product co-occurrence matrix, and the first user similarity is the average of the first similarities between the user and all other users who have taken actions on the product; determine the recommendation priority coefficient of the product for the user based on the first user score, the first product score, the first user similarity, and the first product similarity, wherein the first product score is the average of the scores of all users for the product in the user-product co-occurrence matrix, and the first product similarity is the average of the second similarities between the product and other products with which the user has taken actions; determine the recommendation score of the product for the user based on the preference coefficient, the recommendation priority coefficient, the first interest score, and the second interest score; For each user, the multiple commodities are sorted from large to small according to the recommendation scores of the multiple commodities for the user, a preset number of commodities with the highest sorting order are determined as recommended commodities corresponding to the user, and the recommended commodities are used as sample labels of training samples corresponding to the user.

4. The method according to claim 3, characterized in that The calculation formula of the comprehensive score is: r ui =ω1·b1+ω2·b2+ω3·b3+ω4·b4+ω5·s4 In the formula, r ui represents the comprehensive score of user u for product i, ω1, ω2, ω3, ω4, ω5 are preset weight coefficients, b1, b2, b3, b4, s4 represent the number of views, purchases, searches, evaluations and evaluation scores of user u for product i respectively; The calculation formula of the first similarity is: In the formula, sim(u,v) represents the first similarity between user u and user v, A represents the set of products that both user u and user v have taken actions on, and r uk represents the comprehensive rating of user u for product k, r vk represents the comprehensive rating of user v to product k; The calculation formula of the first interest score is: Where P ui1 represents the first interest rating of user u for product i, N(u) represents the set of N users with the highest first similarity to user u, sim(u,n) represents the first similarity between user u and user n, r ni represents the comprehensive rating of user n on product i; The calculation formula of the second similarity is: In the formula, sim(i,j) represents the second similarity between product i and product j, B represents the set of users who have taken actions on both product i and product j, and r ni represents the comprehensive rating of user n on product i, r nj represents the comprehensive rating of user n on product j; The calculation formula of the second interest score is: Where P ui2 represents the second interest score of user u for product i, N(i) represents the set of N products with the highest second similarity to product i, sim(i,k) represents the second similarity between product i and product k, r uk represents the comprehensive rating of user u on product k; The calculation formula of the preference coefficient is: In the formula, α ui represents the preference coefficient of user u for product i, r ui represents the comprehensive score of user u on product i, C represents the set of users other than user u who have taken actions on product i, sim(u,n) represents the first similarity between user u and user n, and c1 and c2 are preset constants; The calculation formula of the recommended priority coefficient is: In the formula, β ui represents the recommendation priority coefficient of product i to user u, C represents the set of users other than user u who have taken actions on product i, sim(u,n) represents the first similarity between user u and user n, D represents the set of other products other than product i that user u has taken actions on, sim(i,k) represents the second similarity between product i and product k, and r ui represents the comprehensive score of user u on product i, E represents the set of users who have taken actions on product i, r ni represents the comprehensive rating of user n on product i, c3 and c4 are preset constants; The calculation formula of the recommendation score is: S ui =a ui ·b ui ·θ1·P ui1 ·θ2·P ui2 In the formula, S ui represents the recommendation score of product i to user u, α ui represents the preference coefficient of user u for product i, β ui represents the recommendation priority coefficient of product i to user u, P ui1 represents the first interest rating of user u for product i, P ui2 represents the second interest rating of user u for product i, and θ1 and θ2 are preset weight coefficients.

5. The method according to claim 1, characterized in that The target user attribute information includes: the identity identifier, age and gender of the target user. Determining a target commodity recommendation method corresponding to the target user attribute information includes: Querying a commodity recommendation method corresponding to the identity identifier of the target user from a commodity recommendation method list, wherein the commodity recommendation method list stores commodity recommendation methods corresponding to the identity identifiers of different users; If there is a product recommendation method corresponding to the identity identifier of the target user, use the product recommendation method as the target product recommendation method; If there is no product recommendation method corresponding to the identity identifier of the target user, determine a template recommendation method that matches the age and gender of the target user from the product recommendation method list as the target product recommendation method; in response to the target user adjusting the template recommendation method, store the adjusted template recommendation method in the product recommendation method list as the product recommendation method corresponding to the identity identifier of the target user.

6. The method according to claim 3, characterized in that After recommending the target recommended product to the target user based on the target product recommendation method, the method further includes: Obtaining the click-through rate and purchase conversion rate of the target user for the target recommended product, and determining the ratio of the purchase conversion rate to the click-through rate; The target user attribute information, the target user behavior information, the target product attribute information and the target product behavior information are used as a new training sample, and in the process of determining the sample label of the new training sample, when determining the recommendation score of the target recommended product for the target user, the ratio is used as an additional adjustment coefficient to determine the recommendation score of the target recommended product for the target user; The product recommendation model is iteratively trained periodically using new training samples and corresponding sample labels to update model parameters.

7. The method according to claim 1, characterized in that The user attribute information includes at least one of the following: the user's age, gender, address, and completed business; The user behavior information includes at least one of the following: the user's browsing history, purchase history, search history, and evaluation history; The commodity attribute information includes at least one of the following: commodity type, price, brand, and network format; The commodity behavior information includes at least one of the following: commodity browsing records, purchase records, search records, and evaluation records.

8. A commodity recommendation device, characterized in that: include: An acquisition module is used to acquire target user attribute information and target user behavior information of a target user, and target commodity attribute information and target commodity behavior information of all commodities in a commodity pool; An analysis module is used to analyze the target user attribute information, the target user behavior information, the target product attribute information and the target product behavior information by using a pre-trained product recommendation model to obtain target recommended products corresponding to the target user, wherein the training samples for training the product recommendation model include: user attribute information and user behavior information of multiple users, product attribute information and product behavior information of multiple products, and the sample labels for training the product recommendation model are recommended products corresponding to each user determined by the following method: using a collaborative filtering algorithm to analyze the user attribute information, the user behavior information, the product attribute information and the product behavior information to obtain a first similarity between the multiple users and a second similarity between the multiple products, and determining the interest score of each user for each product, the user preference, and the recommendation priority of each product for each user based on the first similarity and the second similarity, and determining the recommended product corresponding to each user based on the interest score, the user preference and the recommendation priority; The recommendation module is used to determine a target commodity recommendation method corresponding to the target user attribute information, and recommend the target recommended commodity to the target user based on the target commodity recommendation method.

9. A computer program product, characterized in that include: A computer program, wherein when the computer program is executed by a processor, the product recommendation method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the commodity recommendation method according to any one of claims 1 to 7 through the computer program.

Images