Product recommendation method, device, computer equipment and medium based on artificial intelligence

By calculating the similarity of user browsing information and training model screening matching, the problem of low recommendation accuracy of existing insurance products is solved, personalized recommendations are achieved, and recommendation accuracy is improved.

CN115080864BActive Publication Date: 2024-11-15CHINA PING AN LIFE INSURANCE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210866909.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-07-21
Publication Date
2024-11-15
Estimated Expiration
2042-07-21

AI Technical Summary

Technical Problem

The existing insurance product recommendation methods rely on manual recommendations, resulting in low recommendation accuracy and difficulty in meeting customers' personalized needs.

Method used

By calculating the similarity of browsing information between users to be recommended and users who have been recommended, determining the reference user, and filtering and matching the purchased products layer by layer. Using the trained product screening and matching model, the most matching insurance products are recommended.

Benefits of technology

It significantly improves the recommendation accuracy of insurance products and ensures that the recommended products are most consistent with customer needs and interests.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115080864B_ABST
    Figure CN115080864B_ABST
Patent Text Reader

Abstract

The present invention is applicable to the field of artificial intelligence technology, and in particular, relates to a product recommendation method, device, computer equipment and medium based on artificial intelligence. The present invention determines a reference user from the recommended users by calculating the similarity between the first browsing information of the user to be recommended and the second browsing information of the recommended user, determines the initial recalled product based on the reference user, clusters all product description information, and arbitrarily selects an initial recalled product from each cluster to form a candidate recalled product. The product description information and the second browsing information of the candidate recalled product are input into a product screening model to determine the selected recalled product, and the product description information and the first browsing information of the selected recalled product are input into a product matching model to make product recommendations to the user to be recommended. By filtering the products purchased by the recommended user layer by layer, the insurance product that best matches the needs and interests of the user to be recommended is determined, which greatly improves the recommendation accuracy of insurance products.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention is applicable to the field of artificial intelligence technology, and in particular relates to a product recommendation method, device, computer equipment and medium based on artificial intelligence. Background Art

[0002] With the development of the insurance industry and the diversification of customer needs, the types of insurance products are increasing, and their complexity is also increasing accordingly, making it more difficult for customers to choose. At present, insurance agents or insurance sales personnel are mainly relied on to recommend products to customers, but the needs of different customer groups are often very different, resulting in poor pertinence and accuracy of the recommendation results, making it difficult to meet customer needs, resulting in great difficulty in recommending suitable insurance products to customers.

[0003] Therefore, in the field of artificial intelligence technology, how to improve the accuracy of insurance product recommendations has become a pressing issue to be addressed. Summary of the invention

[0004] In view of this, the embodiments of the present invention provide a product recommendation method, apparatus, computer device and medium based on artificial intelligence to solve the problem of low accuracy of existing recommendations that rely on manual recommendations.

[0005] In a first aspect, an embodiment of the present invention provides a product recommendation method based on artificial intelligence, wherein the product recommendation comprises:

[0006] Acquire first browsing information of a user to be recommended and second browsing information of a recommended user, calculate information similarity between the first browsing information and the second browsing information, and determine a reference user from the recommended users according to the information similarity;

[0007] Acquire the product purchased by the reference user, and use the product purchased by the reference user as the initial recalled product;

[0008] Obtain product description information of each of the initially recalled products, perform cluster analysis on all product description information to obtain N clusters, and arbitrarily select an initially recalled product from each cluster to form a candidate recalled product, where N is an integer greater than zero;

[0009] Inputting the product description information of the candidate recalled products and the second browsing information into the trained product screening model, determining the screening score of each of the candidate recalled products, and determining the products among the candidate recalled products whose screening scores are not greater than the screening score threshold as selected recalled products;

[0010] The product description information of the selected recalled products and the first browsing information are input into a trained product matching model to determine a matching score for each of the selected recalled products, and the products among the selected recalled products whose matching scores are greater than a matching score threshold are recommended to the user to be recommended.

[0011] In a second aspect, an embodiment of the present invention provides a product recommendation device based on artificial intelligence, the product recommendation device comprising:

[0012] A reference user determination module, configured to obtain first browsing information of a user to be recommended and second browsing information of a recommended user, calculate information similarity between the first browsing information and the second browsing information, and determine a reference user from the recommended users according to the information similarity;

[0013] An initial product determination module, used to obtain the product purchased by the reference user, and use the product purchased by the reference user as the initial recalled product;

[0014] A candidate product determination module, used to obtain product description information of each of the initially recalled products, perform cluster analysis on all product description information to obtain N clusters, and arbitrarily select an initially recalled product from each cluster to form a candidate recalled product, where N is an integer greater than zero;

[0015] a selected product determination module, configured to input the product description information of the candidate recalled products and the second browsing information into a trained product screening model, determine the screening score of each of the candidate recalled products, and determine the product whose screening score is not greater than a screening score threshold among the candidate recalled products as a selected recalled product;

[0016] The product recommendation module is used to input the product description information of the selected recalled products and the first browsing information into the trained product matching model, determine the matching score of each of the selected recalled products, and recommend the products among the selected recalled products whose matching score is greater than the matching score threshold to the user to be recommended.

[0017] In a third aspect, an embodiment of the present invention provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the product recommendation method as described in the first aspect when executing the computer program.

[0018] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the product recommendation method as described in the first aspect is implemented.

[0019] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: by calculating the information similarity between the first browsing information of the user to be recommended and the second browsing information of the recommended user, a reference user is determined from the recommended users, the product purchased by the reference user is used as the initial recalled product, and then all product description information is clustered and analyzed, and an initial recalled product is randomly selected from each cluster to form a candidate recalled product, and then the product description information and the second browsing information of the candidate recalled product are input into the trained product screening model to determine the screening score of each candidate recalled product, and the product whose screening score is not greater than the screening score threshold among the candidate recalled products is determined as the selected recalled product, and finally the product description information and the first browsing information of the selected recalled product are input into the trained product matching model to determine the matching score of each selected recalled product, and the product whose matching score is greater than the matching score threshold among the selected recalled products is recommended to the user to be recommended. By filtering the products purchased by the recommended user layer by layer, the insurance product that best matches the needs and interests of the user to be recommended is determined, and the recommendation accuracy of the insurance product is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 This is a schematic diagram of an application environment of a product recommendation method provided by Embodiment 1 of the present invention;

[0022] Figure 2 is a flowchart of a product recommendation method provided in Embodiment 1 of the present invention;

[0023] Figure 3 is a structural schematic diagram of a product recommendation device provided in Embodiment 2 of the present invention;

[0024] Figure 4 It is a structural diagram of a computer device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0025] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0026] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0027] It should also be understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0028] As used in the present specification and the appended claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]", depending on the context.

[0029] In addition, in the description of the present specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0030] References to "one embodiment" or "some embodiments" etc. described in the present specification mean that one or more embodiments of the present invention include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0031] The embodiments of the present invention can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0032] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0033] It should be understood that the order of execution of the steps in the following embodiments does not imply a precedence of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0034] In order to illustrate the technical solution of the present invention, specific embodiments are provided below for illustration.

[0035] A product recommendation method provided in the first embodiment of the present invention can be applied in the following aspects: Figure 1 In an application environment, a client communicates with a server. The client includes but is not limited to a PDA, a desktop computer, a laptop computer, an ultra-mobile personal computer (UMPC), a netbook, a cloud computer device, a personal digital assistant (PDA), and other computer devices. The server can be implemented by an independent server or a server cluster consisting of multiple servers.

[0036] See also Figure 2 , is a flowchart of a product recommendation method provided in Embodiment 1 of the present invention. The product recommendation method can be applied to Figure 1 In the client, the product recommendation method may include the following steps:

[0037] Step S201, obtaining first browsing information of the user to be recommended and second browsing information of the recommended user, calculating information similarity between the first browsing information and the second browsing information, and determining a reference user from the recommended users according to the information similarity.

[0038] Among them, recommended users are users who have recommended corresponding insurance products, and to-be-recommended users are users who have not recommended corresponding insurance products. It is necessary to make personalized insurance product recommendations based on the needs and interests of to-be-recommended users to meet the needs of customers. The first browsing information and the second browsing information are the process information of users browsing and understanding insurance products before selecting insurance products, including browsing categories, browsing time and browsing order. The browsing information of users can effectively reflect the interests and needs of users.

[0039] Therefore, the first browsing information of the user to be recommended and the second browsing information of the recommended user can be obtained, and the information similarity between the first browsing information and the second browsing information can be calculated. When the information similarity between the first browsing information and the second browsing information is high, it means that the interests and needs of the user to be recommended and the recommended user are highly similar. Therefore, the recommended user corresponding to the second browsing information similar to the first browsing information is used as a reference user to provide reference information for the personalized insurance product recommendation for the recommended user, which can effectively improve the accuracy of insurance product recommendation.

[0040] Optionally, calculating the information similarity between the first browsing information and the second browsing information, and determining the reference user from the recommended users according to the information similarity includes:

[0041] Convert the first browsing information into a first browsing vector according to the word vector technology, and convert the second browsing information into a second browsing vector according to the word vector technology;

[0042] Calculating the browsing vector similarity between the first browsing vector and the second browsing vector as the information similarity between the first browsing information and the second browsing information;

[0043] The relationship between the information similarity and a preset information similarity threshold is determined. If the information similarity is greater than the information similarity threshold, the recommended user corresponding to the information similarity is determined as a reference user.

[0044] The first browsing information and the second browsing information are text information. Before calculating the information similarity, the text information needs to be represented as a vector that can express the text semantics. Therefore, the first browsing information and the second browsing information are first vectorized using word vector technology to obtain a first browsing vector corresponding to the first browsing information and a second browsing vector corresponding to the second browsing information. The word vector technology includes, for example, One-hot Encoding technology and Word2Vec technology.

[0045] Then, the browsing vector similarity between the first browsing vector and the second browsing vector is calculated. The higher the browsing vector similarity, the more similar the first browsing vector is to the second browsing vector, and the more similar the corresponding first browsing information and the second browsing information are. Therefore, the browsing vector similarity between the first browsing vector and the second browsing vector is used as the information similarity between the first browsing information and the second browsing information.

[0046] In order to filter out users that can be used as references from the recommended users, the obtained information similarity is compared with the preset information similarity threshold. If the information similarity is greater than the information similarity threshold, it indicates that the interests and needs of the recommended user corresponding to the information similarity for insurance products are similar to the interests and needs of the user to be recommended for insurance products. In this case, the recommended user can be used as a reference user to provide reference information for the personalized insurance product recommendation of the recommended user. If the information similarity is not greater than the information similarity threshold, it indicates that the interests and needs of the recommended user corresponding to the information similarity for insurance products are similar to the interests and needs of the user to be recommended for insurance products. In this case, the recommended user is not used as a reference user, thereby effectively improving the accuracy of the recommendation of insurance products. The information similarity threshold can be preset according to actual conditions. In this embodiment, the information similarity threshold is preset to 0.8 according to actual conditions to ensure the credibility of the reference user.

[0047] Step S202, obtaining the products purchased by the reference user, and taking the products purchased by the reference user as the initial recalled products.

[0048] Among them, the reference user's interests and demands for insurance products are highly similar to the interests and demands of the user to be recommended for insurance products. Therefore, this embodiment obtains the products purchased by the reference user and uses the products purchased by the reference user as initial recalled products for reference when making personalized product recommendations for the user to be recommended.

[0049] Step S203, obtaining product description information of each initially recalled product, performing cluster analysis on all product description information to obtain N clusters, and arbitrarily selecting an initially recalled product from each cluster to form a candidate recalled product.

[0050] Among them, the product description information includes the insurance period, insurance amount, and insured age of the insurance product, and introduces the relevant information of the corresponding insurance product to the insured person in the form of text content. Since the initial recall products are products purchased by the reference user, and the reference user and the recommended user have a high degree of similarity in the interests and needs of insurance products, the initial recall products contain more consistent or similar insurance products. When making targeted insurance product recommendations to the recommended user, it is necessary to screen the consistent or similar insurance products in the initial recall products to ensure that the differences between the initial recall products retained after screening are large, so as to provide more different insurance product options based on the interests and needs of the recommended user, while reducing the amount of calculation, so as to improve the accuracy of insurance product recommendations and improve the efficiency of recommendations.

[0051] Specifically, after obtaining the product description information of each initially recalled product, cluster analysis is performed on all product description information according to the clustering algorithm to obtain N clusters. The similarity between the initially recalled products belonging to the same cluster is relatively high. Therefore, for each cluster, one of the initially recalled products is randomly selected as a candidate recalled product to complete the screening of consistent or similar insurance products in the initially recalled products, so as to ensure that the differences between the initially recalled products retained after screening are relatively large, so as to simultaneously ensure the accuracy of insurance product recommendations and improve recommendation efficiency. Wherein, N is an integer greater than zero.

[0052] Step S204, input the product description information and the second browsing information of the candidate recalled products into the trained product screening model, determine the screening score of each candidate recalled product, and determine the products among the candidate recalled products whose screening scores are not greater than the screening score threshold as selected recalled products.

[0053] Among them, the candidate recalled products are obtained from massive insurance products through preliminary screening of recommended users and products purchased by recommended users, and are used to make personalized product recommendations to the recommended users. When the number of candidate recalled products is large, the recommendation efficiency for the recommended users will be seriously reduced. Therefore, this embodiment further screens the candidate recalled products to greatly reduce the number of recalled products, thereby improving the recommendation efficiency of insurance products.

[0054] Specifically, the second browsing information is the process information of the recommended user browsing and understanding the insurance product before selecting the insurance product, which effectively shows the interests and needs of the recommended user. The candidate recall product is part of the products purchased by the reference user among the recommended users. The product description information of the candidate recall product can verify the credibility of judging the interests and needs of the recommended user for the insurance product according to the second browsing information. Therefore, the product description information and the second browsing information of the candidate recall product are input into the trained product screening model to obtain the screening score of each candidate recall product. The larger the screening score, the greater the probability that the candidate recall product is screened out. The screening score of each candidate recall product is further compared with the screening score threshold. If the screening score is greater than the screening score threshold, the corresponding candidate recall product is screened out. If the screening score is not greater than the screening score threshold, the corresponding candidate recall product is retained. Finally, the products among the candidate recall products whose screening scores are not greater than the screening score threshold are determined as selected recall products, and personalized product recommendations are made to the recommended users as candidate recall products after quantity downgrade. The screening score threshold may be preset according to actual conditions. In this embodiment, the screening score threshold is preset to 0.3 according to actual conditions to ensure the effectiveness of quantity downgrading for candidate recalled products.

[0055] Optionally, the product screening model includes a product coarse screen model and a product fine screen model:

[0056] The product rough screening model is used to input the product description information and second browsing information of the candidate recalled products, and output the dimension reduction features and rough screening scores of the candidate recalled products;

[0057] The product fine screening model is used to input the product description information, second browsing information and dimension reduction features of the candidate recalled products, and output the fine screening scores of the candidate recalled products;

[0058] The screening score of each candidate recall product is determined based on the coarse screening score and the fine screening score, and the selected recall products are determined from the candidate recall products based on the screening score.

[0059] Among them, the product screening model is divided into a product coarse screening model and a product fine screening model, which conducts coarse screening and fine screening on the candidate recall products to improve the screening accuracy of the product screening model, effectively reduce the number of candidate recall products, and improve the recommendation efficiency for insurance products.

[0060] Specifically, the product coarse screening model is used to input the product description information and the second browsing information of the candidate recalled products, perform feature extraction on the product description information and the second browsing information, and output the dimensionality reduction features and coarse screening scores of the candidate recalled products. The product fine screening model is used to input the product description information, the second browsing information and the dimensionality reduction features of the candidate recalled products, and finally output the fine screening scores of the candidate recalled products. The final screening score of each candidate recalled product is determined based on the coarse screening score and the fine screening score. The higher the screening score, the greater the probability that the corresponding candidate recalled product will be screened out. Therefore, the candidate recalled products can be screened according to the screening score, and the selected recalled products can be determined from the candidate recalled products to improve the recommendation efficiency of insurance products.

[0061] Optionally, the product coarse screening model includes a coarse screening encoder, a coarse screening embedding layer and a coarse screening fully connected layer, with the product description information of the sample recalled product and the corresponding second browsing information as training samples, and a preset coarse screening score as a training label;

[0062] The training process of the product coarse screen model includes:

[0063] Input the product description information of the sample recalled product and the corresponding second browsing information into the coarse screening encoder for feature extraction to obtain the sample coarse screening features;

[0064] Input the sample coarse screening features into the coarse screening embedding layer to perform feature dimension reduction, and obtain the sample dimension reduction features;

[0065] Input the sample dimension reduction features into the coarse screening fully connected layer to obtain the sample coarse screening score;

[0066] The first loss function is calculated according to the sample coarse screening score and the preset coarse screening score, and the parameters of the coarse screening encoder, coarse screening embedding layer and coarse screening fully connected layer are reversely corrected according to the gradient descent method until the first loss function converges to obtain the trained product coarse screening model.

[0067] Among them, the training samples are the product description information and corresponding second browsing information of a large number of sample recalled products, and the training labels are preset coarse screening scores, which are used to evaluate the sample coarse screening scores obtained by the product coarse screening model, so that the product coarse screening model can be trained according to the evaluation results to obtain a trained product coarse screening model.

[0068] During the training process of the product coarse screening model, the product description information of the sample recalled product and the corresponding second browsing information are first input into the coarse screening encoder for feature extraction to obtain the sample coarse screening features. The sample coarse screening features are used to represent the correlation between the product description information of the sample recalled product and the corresponding second browsing information, so as to verify the credibility of the recommended user's interest and demand for insurance products based on the second browsing information according to the product description information of the candidate recalled product; the sample coarse screening features are further input into the coarse screening embedding layer for feature dimensionality reduction to obtain the sample dimensionality reduction features, and finally the sample dimensionality reduction features are input into the coarse screening fully connected layer to obtain the sample coarse screening score, which is used to represent the probability of the corresponding candidate recalled product being screened out in the coarse screening dimensionality reduction process.

[0069] Then, the first loss function is calculated according to the sample coarse screening score and the preset coarse screening score. When the first loss function is small, it means that the gap between the sample coarse screening score obtained according to the product coarse screening model and the actual preset coarse screening score is small, indicating that the coarse screening accuracy of the product coarse screening model is high. When the first loss function is large, it means that the coarse screening accuracy of the product coarse screening model is low. It is necessary to reversely correct the parameters of the coarse screening encoder, coarse screening embedding layer and coarse screening fully connected layer according to the gradient descent method until the first loss function converges to obtain the trained product coarse screening model to improve the coarse screening accuracy of the product coarse screening model.

[0070] Optionally, the product fine screening model includes a fine screening encoder and a fine screening fully connected layer, using product description information, second browsing information and sample dimension reduction features of sample recalled products as training samples, and using preset fine screening scores as training labels;

[0071] The training process of the product fine sieve model includes:

[0072] Input the product description information, the second browsing information and the sample dimension reduction features of the sample recalled product into the fine screening encoder for feature extraction to obtain the sample fine screening features;

[0073] The sample fine screening features and the sample dimensionality reduction features are fused to obtain the sample fusion features;

[0074] Input the sample fusion features into the fine screening fully connected layer to obtain the sample fine screening score;

[0075] The second loss function is calculated according to the sample fine screening score and the preset fine screening score, and the parameters of the fine screening encoder and the fine screening fully connected layer are reversely corrected according to the gradient descent method until the second loss function converges to obtain the trained product fine screening model.

[0076] Among them, the training samples are product description information of a large number of sample recalled products, the corresponding second browsing information and the sample dimensionality reduction features obtained according to the product coarse sieve model, and the training labels are preset fine screening scores, which are used to evaluate the sample fine screening scores obtained by the product fine sieve model, so that the product fine sieve model can be trained according to the evaluation results to obtain a trained product fine sieve model.

[0077] During the training process of the product fine screening model, the product description information, the second browsing information and the sample dimensionality reduction features of the sample recalled product are first input into the fine screening encoder for feature extraction to obtain the sample fine screening features. The sample fine screening features are used to represent the correlation between the product description information of the sample recalled product and the corresponding second browsing information, so as to verify the credibility of the recommended user's interest and demand for insurance products based on the second browsing information according to the product description information of the candidate recalled product, and the sample fine screening features and the sample dimensionality reduction features are feature fused to obtain the sample fusion features to judge the feasibility of dimensionality reduction for the candidate recalled product. Finally, the sample fusion features are input into the fine screening fully connected layer to obtain the sample fine screening score, which is used to represent the probability of the corresponding candidate recalled product being screened out in the fine screening dimensionality reduction process.

[0078] Then, the second loss function is calculated according to the sample fine screening score and the preset fine screening score. When the second loss function is small, it means that the gap between the sample fine screening score obtained according to the product fine screening model and the actual preset fine screening score is small, indicating that the fine screening accuracy of the product fine screening model is high. When the second loss function is large, it means that the fine screening accuracy of the product fine screening model is low, and it is necessary to reversely correct the parameters of the fine screening encoder and the fine screening fully connected layer according to the gradient descent method until the second loss function converges to obtain the trained product fine screening model to improve the fine screening accuracy of the product fine screening model.

[0079] Optionally, a screening score for each candidate recall product is determined based on the coarse screening score and the fine screening score, and selected recall products are determined from the candidate recall products based on the screening score, including:

[0080] Inputting the coarse screening score, the preset first score weight, the fine screening score, and the preset second score weight into a preset score model to determine the screening score of each candidate recall product;

[0081] The relationship between the screening score and the screening score threshold is determined. If the screening score is not greater than the screening score threshold, the candidate recall product is determined to be a selected recall product.

[0082] Among them, the coarse screening score is used to indicate the probability of the corresponding candidate recalled product being screened out in the coarse screening and dimensionality reduction process, and the fine screening score is used to indicate the probability of the corresponding candidate recalled product being screened out in the fine screening and dimensionality reduction process. The coarse screening score and the fine screening score, as well as the first score weight corresponding to the coarse screening score and the second score weight corresponding to the fine screening score are combined to comprehensively judge the probability of the corresponding candidate recalled product being screened out in the coarse screening and dimensionality reduction process, thereby determining the selected recalled products among the candidate recalled products, so as to greatly reduce the number of candidate recalled products, and make personalized product recommendations to the recommended users based on the selected recalled products, so as to improve the recommendation efficiency and accuracy of insurance products.

[0083] Specifically, the coarse screening score, the preset first score weight, the fine screening score, and the preset second score weight are input into the preset scoring model to determine the screening score of each candidate recall product. The larger the screening score, the greater the probability that the corresponding candidate recall product will be screened out in the fine screening and dimensionality reduction process. Therefore, the relationship between the screening score and the screening score threshold is further judged. If the screening score is greater than the screening score threshold, the corresponding candidate recall product will be screened out. If the screening score is not greater than the screening score threshold, the corresponding candidate recall product will be retained, and the corresponding candidate recall product will be determined as a selected recall product, so as to make personalized product recommendations to the recommended users. Among them, the preset first score weight and the preset second score weight can be preset according to actual conditions, and are used to indicate the contribution of the corresponding coarse screening scores and fine screening scores in calculating the probability of the corresponding candidate recalled products being screened out. Since the fine screening score is obtained by further screening the candidate recalled products on the basis of the coarse screening score, the contribution of the fine screening score in calculating the probability of the corresponding candidate recalled products being screened out is greater than the contribution of the coarse screening score. Therefore, when presetting the weights according to actual conditions, it is necessary to ensure that the preset second score weight is greater than the preset first score weight.

[0084] For example, the number of candidate recall products is recorded as N, where N is an integer greater than 1, and the rough screening score of the i-th (i=1, 2, ..., N) candidate recall product is recorded as C i , the fine screening score of the i-th candidate recalled product is recorded as X i According to the actual situation, the first score weight is preset as α1 = 0.2, the second score weight is preset as α2 = 0.8, and the score model is preset as the product of the coarse screening score and the first score weight, plus the product of the fine screening score and the second score weight, to obtain the corresponding screening score. Therefore, the screening score S of the i-th candidate recall product is i for:

[0085] Si =α1C i +α2X i

[0086] In the formula, α1 is the preset first score weight, C i is the rough screening score of the i-th candidate recalled product, α2 is the preset second score weight, X i is the fine screening score of the i-th candidate recalled product.

[0087] Then the screening scores S1, S2, ..., S of N candidate recalled products can be obtained. N , and further normalize the N screening scores to obtain the normalized screening scores S1v, S2′, ..., S N ′, so that the normalized screening score ranges from [0, 1], and the screening score threshold is preset as S according to the actual situation. o =0.3, if the normalized screening score of the i-th candidate recalled product is S i ′>S0, the i-th candidate recalled product is screened out; if the normalized screening score of the i-th candidate recalled product is S i ′≤S0, the i-th candidate recalled product is retained and used as a selected recalled product for personalized product recommendation to the recommended user.

[0088] Step S205, input the product description information and the first browsing information of the selected recalled products into the trained product matching model, determine the matching score of each selected recalled product, and recommend the products with matching scores greater than the matching score threshold among the selected recalled products to the recommended users.

[0089] Among them, the selected recalled products are obtained from a large number of insurance products through multiple screening of recommended users and products purchased by recommended users, which can effectively show the interests and needs of recommended users and are used to make personalized product recommendations for recommended users. The first browsing information is the process information of the recommended user browsing and understanding the insurance product before selecting the insurance product, which effectively shows the interests and needs of the recommended user. Therefore, in order to improve the recommendation accuracy of insurance products, the product description information and the first browsing information of the selected recalled products are input into the trained product matching model to obtain the matching score of each selected recalled product, which is used to indicate the matching degree between the selected recalled products of the recommended user and the first browsing information of the recommended user, so as to recommend the products with matching scores greater than the matching score threshold in the selected recalled products to the recommended user, so as to complete the targeted insurance product recommendation for the recommended user according to the first browsing information of the recommended user. Among them, the matching score threshold can be preset according to the actual situation. In this embodiment, the matching score threshold is preset as 0.75 according to the actual situation to ensure the recommendation accuracy of insurance products.

[0090] Optionally, the product matching model includes a matching encoder and a matching fully connected layer, using product description information and corresponding first browsing information of sample selected recalled products as training samples, and using a preset matching score as a training label;

[0091] The training process of the product matching model includes:

[0092] Inputting the product description information of the sample selected recalled product and the corresponding first browsing information into the matching encoder for feature extraction to obtain the sample matching feature;

[0093] Input the sample matching features into the matching fully connected layer to obtain the sample matching score;

[0094] The third loss function is calculated according to the sample matching score and the preset matching score, and the parameters of the matching encoder and the matching fully connected layer are reversely corrected according to the gradient descent method until the third loss function converges to obtain a trained product matching model.

[0095] Among them, the training samples are product description information and corresponding first browsing information of a large number of sample selected recalled products, and the training labels are preset matching scores, which are used to evaluate the sample matching scores obtained by the product matching model, so that the product matching model can be trained according to the evaluation results to obtain a trained product matching model.

[0096] During the training process of the product matching model, the product description information of the sample selected recalled products and the corresponding first browsing information are first input into the matching encoder for feature extraction to obtain the sample matching features, which are used to represent the matching degree between the selected recalled products of the recommended users and the first browsing information of the users to be recommended. The sample matching features are then input into the matching fully connected layer to obtain the sample matching score. The larger the sample matching score, the higher the priority of the product in being recommended to the users to be recommended.

[0097] Then, the third loss function is calculated based on the sample matching score and the preset matching score. When the third loss function is small, it means that the gap between the sample matching score obtained according to the product matching model and the actual preset matching score is small, indicating that the matching accuracy of the product matching model is high. When the third loss function is large, it means that the matching accuracy of the product matching model is low, and it is necessary to reversely correct the parameters of the matching encoder and the matching fully connected layer according to the gradient descent method until the third loss function converges to obtain the trained product matching model to improve the matching accuracy of the product matching model.

[0098] The embodiment of the present invention determines a reference user from the recommended users by calculating the information similarity between the first browsing information of the user to be recommended and the second browsing information of the recommended user, takes the product purchased by the reference user as the initial recalled product, and then performs cluster analysis on all product description information, selects an initial recalled product from each cluster at random to form a candidate recalled product, and then inputs the product description information and the second browsing information of the candidate recalled product into a trained product screening model to determine the screening score of each candidate recalled product, and determines that the product whose screening score is not greater than the screening score threshold among the candidate recalled products is a selected recalled product, and finally inputs the product description information and the first browsing information of the selected recalled product into a trained product matching model to determine the matching score of each selected recalled product, and recommends the product whose matching score is greater than the matching score threshold among the selected recalled products to the user to be recommended. By filtering the products purchased by the recommended user layer by layer, the insurance product that best matches the needs and interests of the user to be recommended is determined, and the recommendation accuracy of the insurance product is greatly improved.

[0099] Corresponding to the product recommendation method in the above embodiment, Figure 3 A structural block diagram of a product recommendation device provided in the second embodiment of the present invention is given. For the convenience of description, only the parts related to the embodiment of the present invention are shown.

[0100] See also Figure 3 , the recommended devices for this product include:

[0101] A reference user determination module 31 is used to obtain first browsing information of a user to be recommended and second browsing information of a recommended user, calculate information similarity between the first browsing information and the second browsing information, and determine a reference user from the recommended users according to the information similarity;

[0102] An initial product determination module 32 is used to obtain the product purchased by the reference user and use the product purchased by the reference user as the initial recalled product;

[0103] The candidate product determination module 33 is used to obtain product description information of each initially recalled product, perform cluster analysis on all product description information to obtain N clusters, and select an initially recalled product from each cluster to constitute a candidate recalled product, where N is an integer greater than zero;

[0104] A selected product determination module 34 is used to input the product description information of the candidate recalled products and the second browsing information into the trained product screening model, determine the screening score of each candidate recalled product, and determine the product whose screening score is not greater than the screening score threshold among the candidate recalled products as the selected recalled product;

[0105] The product recommendation module 35 is used to input the product description information and the first browsing information of the selected recalled products into the trained product matching model, determine the matching score of each selected recalled product, and recommend the products with matching scores greater than the matching score threshold among the selected recalled products to the recommended users.

[0106] Optionally, the reference user determination module 31 includes:

[0107] A browsing vector conversion submodule, used to convert the first browsing information into a first browsing vector according to the word vector technology, and to convert the second browsing information into a second browsing vector according to the word vector technology;

[0108] An information similarity calculation submodule, used to calculate the browsing vector similarity between the first browsing vector and the second browsing vector as the information similarity between the first browsing information and the second browsing information;

[0109] The reference user determination submodule is used to determine the relationship between the information similarity and a preset information similarity threshold. If the information similarity is greater than the information similarity threshold, the recommended user corresponding to the information similarity is determined as the reference user.

[0110] Optionally, the selected product determination module 34 includes:

[0111] A coarse screening score determination submodule is used for a product coarse screening submodel to input product description information and second browsing information of a candidate recalled product, and output a dimension reduction feature and a coarse screening score of the candidate recalled product;

[0112] A fine screening score determination submodule is used for a product fine screening model to input product description information, second browsing information and dimension reduction features of a candidate recalled product, and output a fine screening score of the candidate recalled product;

[0113] The selected product determination submodule is used to determine the screening score of each candidate recalled product according to the coarse screening score and the fine screening score, and determine the selected recalled products from the candidate recalled products according to the screening score.

[0114] Optionally, the above-mentioned coarse screening score determination submodule includes:

[0115] A coarse screening feature extraction unit, used for inputting the product description information of the sample recalled product and the corresponding second browsing information into the coarse screening encoder for feature extraction to obtain the sample coarse screening feature;

[0116] A dimension reduction feature determination unit is used to input the sample coarse screening feature into the coarse screening embedding layer to perform feature dimension reduction to obtain the sample dimension reduction feature;

[0117] A coarse screening score determination unit is used to input the sample dimension reduction features into the coarse screening fully connected layer to obtain the sample coarse screening score;

[0118] The first parameter correction unit is used to calculate the first loss function according to the sample coarse screening score and the preset coarse screening score, and reversely correct the parameters of the coarse screening encoder, the coarse screening embedding layer and the coarse screening fully connected layer according to the gradient descent method until the first loss function converges to obtain a trained product coarse screening model.

[0119] Optionally, the above-mentioned fine screening score determination submodule includes:

[0120] A fine screening feature extraction unit, used for inputting the product description information, the second browsing information and the sample dimension reduction feature of the sample recalled product into the fine screening encoder for feature extraction to obtain the sample fine screening feature;

[0121] A feature fusion unit is used to fuse the sample fine screening feature and the sample dimensionality reduction feature to obtain a sample fusion feature;

[0122] A fine screening score determination unit is used to input the sample fusion features into the fine screening fully connected layer to obtain the sample fine screening score;

[0123] The second parameter correction unit is used to calculate the second loss function according to the sample fine screening score and the preset fine screening score, and reversely correct the parameters of the fine screening encoder and the fine screening fully connected layer according to the gradient descent method until the second loss function converges to obtain a trained product fine screening model.

[0124] Optionally, the above-mentioned selected product determination submodule includes:

[0125] A screening score determination unit, used to input the coarse screening score, the preset first score weight, the fine screening score, and the preset second score weight into a preset score model to determine the screening score of each candidate recall product;

[0126] The selected product determination unit is used to determine the relationship between the screening score and the screening score threshold. If the screening score is not greater than the screening score threshold, the candidate recall product is determined to be a selected recall product.

[0127] Optionally, the product recommendation module 35 includes:

[0128] A matching feature extraction submodule, used for inputting the product description information of the sample selected recalled product and the corresponding first browsing information into a matching encoder for feature extraction to obtain a sample matching feature;

[0129] A matching score determination submodule is used to input the sample matching features into the matching fully connected layer to obtain the sample matching score;

[0130] The third parameter correction submodule is used to calculate the third loss function according to the sample matching score and the preset matching score, and reversely correct the parameters of the matching encoder and the matching fully connected layer according to the gradient descent method until the third loss function converges to obtain a trained product matching model.

[0131] It should be noted that the information interaction, execution process and other contents between the above modules are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0132] Figure 4 This is a schematic diagram of the structure of a computer device provided in Embodiment 3 of the present invention. Figure 4 As shown, the computer device of this embodiment includes: at least one processor ( Figure 4 Only one is shown), a memory, and a computer program stored in the memory and executable on at least one processor, wherein when the processor executes the computer program, the steps in any of the above-mentioned product recommendation method embodiments are implemented.

[0133] The computer device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that Figure 4 This is merely an example of a computer device and does not constitute a limitation on the computer device. The computer device may include more or fewer components than those shown in the figure, or a combination of certain components, or different components. For example, it may also include a network interface, a display screen, and an input device.

[0134] The processor may be a CPU, or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0135] The memory includes a readable storage medium, an internal memory, etc., wherein the internal memory may be the memory of a computer device, and the internal memory provides an environment for the operation of an operating system and computer-readable instructions in the readable storage medium. The readable storage medium may be a hard disk of a computer device, and in other embodiments may also be an external storage device of a computer device, for example, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on a computer device. Further, the memory may also include both an internal storage unit of a computer device and an external storage device. The memory is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as program codes of computer programs, etc. The memory may also be used to temporarily store data that has been output or is to be output.

[0136] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention. The specific working process of the units and modules in the above-mentioned device can refer to the corresponding process in the above-mentioned method embodiment, which will not be repeated here. 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 such understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned method embodiment can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device capable of carrying computer program code, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0137] The present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by a computer program product. When the computer program product runs on a computer device, the computer device can implement the steps in the above-mentioned method embodiment when executing.

[0138] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0139] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0140] In the embodiments provided by the present invention, it should be understood that the disclosed devices / computer equipment and methods can be implemented in other ways. For example, the device / computer equipment embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as 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 devices or units, which can be electrical, mechanical or other forms.

[0141] 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 on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0142] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A product recommendation method based on artificial intelligence, characterized in that: The product recommendation method comprises: Acquire first browsing information of a user to be recommended and second browsing information of a recommended user, calculate information similarity between the first browsing information and the second browsing information, and determine a reference user from the recommended users according to the information similarity; Acquire the products purchased by the reference user, and use the products purchased by the reference user as initial recalled products; Obtain product description information of each of the initially recalled products, perform cluster analysis on all product description information to obtain N clusters, and arbitrarily select an initially recalled product from each cluster to form a candidate recalled product, where N is an integer greater than zero; Inputting the product description information of the candidate recalled products and the second browsing information into the trained product screening model, determining the screening score of each of the candidate recalled products, and determining the products among the candidate recalled products whose screening scores are not greater than the screening score threshold as selected recalled products; Inputting the product description information of the selected recalled products and the first browsing information into a trained product matching model, determining a matching score of each of the selected recalled products, and recommending the products among the selected recalled products whose matching scores are greater than a matching score threshold to the to-be-recommended user; The product screening model includes a product coarse sieve model and a product fine sieve model: The product rough screening model is used to input the product description information of the candidate recalled product and the second browsing information, and output the dimension reduction features and rough screening scores of the candidate recalled product; The product fine screening model is used to input the product description information of the candidate recalled product, the second browsing information and the dimensionality reduction feature, and output the fine screening score of the candidate recalled product; Determine a screening score for each of the candidate recalled products according to the coarse screening score and the fine screening score, and determine a selected recalled product from the candidate recalled products according to the screening score; The product coarse screening model includes a coarse screening encoder, a coarse screening embedding layer and a coarse screening fully connected layer, takes the product description information of the sample recalled product and the corresponding second browsing information as training samples, and takes the preset coarse screening score as the training label; The training process of the product coarse screen model includes: Inputting the product description information of the sample recalled product and the corresponding second browsing information into the coarse screening encoder for feature extraction to obtain sample coarse screening features; Inputting the sample coarse screening features into the coarse screening embedding layer to perform feature dimension reduction to obtain sample dimension reduction features; Inputting the sample dimension reduction features into the coarse screening fully connected layer to obtain the sample coarse screening score; A first loss function is calculated according to the sample coarse screening score and the preset coarse screening score, and the parameters of the coarse screening encoder, the coarse screening embedding layer and the coarse screening fully connected layer are reversely corrected according to the gradient descent method until the first loss function converges to obtain a trained product coarse screening model.

2. The product recommendation method according to claim 1, characterized in that: The calculating the information similarity between the first browsing information and the second browsing information, and determining a reference user from the recommended users according to the information similarity, includes: Convert the first browsing information into a first browsing vector according to the word vector technology, and convert the second browsing information into a second browsing vector according to the word vector technology; calculating a browsing vector similarity between the first browsing vector and the second browsing vector as information similarity between the first browsing information and the second browsing information; The relationship between the information similarity and a preset information similarity threshold is determined, and if the information similarity is greater than the information similarity threshold, the recommended user corresponding to the information similarity is determined as a reference user.

3. The product recommendation method according to claim 1, characterized in that: The product fine screening model includes a fine screening encoder and a fine screening fully connected layer, takes the product description information of the sample recalled product, the second browsing information and the sample dimensionality reduction features as training samples, and takes the preset fine screening scores as training labels; The training process of the product fine sieve model includes: Inputting the product description information of the sample recalled product, the second browsing information and the sample dimension reduction feature into the fine screening encoder for feature extraction to obtain the sample fine screening feature; Performing feature fusion on the sample fine screening feature and the sample dimensionality reduction feature to obtain a sample fusion feature; Inputting the sample fusion features into the fine screening fully connected layer to obtain the sample fine screening score; A second loss function is calculated based on the sample fine screening score and the preset fine screening score, and the parameters of the fine screening encoder and the fine screening fully connected layer are reversely corrected according to the gradient descent method until the second loss function converges to obtain a trained product fine screening model.

4. The product recommendation method according to claim 1, characterized in that: Determining the screening score of each of the candidate recalled products according to the coarse screening score and the fine screening score, and determining the selected recalled products from the candidate recalled products according to the screening score, comprises: Inputting the coarse screening score, the preset first score weight, the fine screening score, and the preset second score weight into a preset score model to determine the screening score of each of the candidate recalled products; The relationship between the screening score and the screening score threshold is determined. If the screening score is not greater than the screening score threshold, the candidate recalled product is determined to be a selected recalled product.

5. The product recommendation method according to claim 1, characterized in that: The product matching model includes a matching encoder and a matching fully connected layer, using the product description information and the corresponding first browsing information of the sample selected recalled products as training samples, and using the preset matching scores as training labels; The training process of the product matching model includes: Inputting the product description information of the sample selected recalled product and the corresponding first browsing information into the matching encoder for feature extraction to obtain sample matching features; Inputting the sample matching features into the matching fully connected layer to obtain a sample matching score; A third loss function is calculated according to the sample matching score and the preset matching score, and the parameters of the matching encoder and the matching fully connected layer are reversely corrected according to the gradient descent method until the third loss function converges to obtain a trained product matching model.

6. A product recommendation device based on artificial intelligence, characterized in that: The product recommendation device comprises: A reference user determination module, configured to obtain first browsing information of a user to be recommended and second browsing information of a recommended user, calculate information similarity between the first browsing information and the second browsing information, and determine a reference user from the recommended users according to the information similarity; An initial product determination module, used to obtain the product purchased by the reference user, and use the product purchased by the reference user as the initial recalled product; A candidate product determination module, used to obtain product description information of each of the initially recalled products, perform cluster analysis on all product description information to obtain N clusters, and arbitrarily select an initially recalled product from each cluster to form a candidate recalled product, where N is an integer greater than zero; a selected product determination module, configured to input the product description information of the candidate recalled products and the second browsing information into a trained product screening model, determine the screening score of each of the candidate recalled products, and determine that the product whose screening score is not greater than a screening score threshold among the candidate recalled products is a selected recalled product; A product recommendation module, configured to input the product description information of the selected recalled products and the first browsing information into a trained product matching model, determine a matching score of each of the selected recalled products, and recommend the products among the selected recalled products whose matching scores are greater than a matching score threshold to the to-be-recommended user; The product screening model includes a product coarse sieve model and a product fine sieve model: The product rough screening model is used to input the product description information of the candidate recalled product and the second browsing information, and output the dimension reduction features and rough screening scores of the candidate recalled product; The product fine screening model is used to input the product description information of the candidate recalled product, the second browsing information and the dimensionality reduction feature, and output the fine screening score of the candidate recalled product; Determine a screening score for each of the candidate recalled products according to the coarse screening score and the fine screening score, and determine a selected recalled product from the candidate recalled products according to the screening score; The product coarse screening model includes a coarse screening encoder, a coarse screening embedding layer and a coarse screening fully connected layer, takes the product description information of the sample recalled product and the corresponding second browsing information as training samples, and takes the preset coarse screening score as the training label; The training process of the product coarse screen model includes: Inputting the product description information of the sample recalled product and the corresponding second browsing information into the coarse screening encoder for feature extraction to obtain sample coarse screening features; Inputting the sample coarse screening features into the coarse screening embedding layer to perform feature dimension reduction to obtain sample dimension reduction features; Inputting the sample dimension reduction features into the coarse screening fully connected layer to obtain the sample coarse screening score; A first loss function is calculated according to the sample coarse screening score and the preset coarse screening score, and the parameters of the coarse screening encoder, the coarse screening embedding layer and the coarse screening fully connected layer are reversely corrected according to the gradient descent method until the first loss function converges to obtain a trained product coarse screening model.

7. A computer device, characterized in that: The computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the product recommendation method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the product recommendation method according to any one of claims 1 to 5 is implemented.