Product recommendation method and device, equipment and storage medium

Abstract

The invention relates to the technical field of product recommendation and artificial intelligence, and provides a product recommendation method and device, equipment and a storage medium, and the method comprises the steps: obtaining a browsing record of a to-be-responded user browsing a product; according to the browsing record of the to-be-responded user, determining the user intention feature of the to-be-responded user, the interaction intention of the to-be-responded user to the product and the product feature; determining a first fusion feature corresponding to the to-be-responded user according to the interaction intention of the to-be-responded user and the product feature of the product interacted by the to-be-responded user; determining a second fusion feature corresponding to the product according to the interaction intention of the to-be-responded user to the product, the user intention feature and the product feature; and determining whether the product is a target recommended product according to the first fusion feature and the second fusion feature. By taking the field of medical health or finance as an example, whether the product is the target recommended product can be accurately determined according to the fusion feature corresponding to the to-be-responded user interaction product and the fusion feature corresponding to the user intention feature associated with the product.

Description

Technical Field

[0001] This application relates to the fields of product recommendation and artificial intelligence technology, and in particular to a product recommendation method, apparatus, device and storage medium. Background Technology

[0002] As the group directly communicating with users, call center agents bear the key tasks of handling user inquiries about products and facilitating business cooperation. In the healthcare context, agents can recommend various business products such as recuperation services, health check-up packages, and nursing services to users through intelligent response systems. As the business scope of medical, financial, and other institutions expands, the product content inquired about by users has become more complex and diverse. Therefore, agents face numerous challenges in handling user inquiries and recommending products.

[0003] In related technologies, when agents recommend products to users, they mostly rely on their own experience and basic understanding of the products. This makes it difficult to recommend products that meet the user's needs in a timely and accurate manner, which not only reduces the user experience but also leads to low work efficiency for agents. Summary of the Invention

[0004] The main objective of this application is to provide a product recommendation method, apparatus, device, and storage medium that can analyze the impact of products on users and the impact of users on products from two different perspectives, so as to accurately recommend products that meet the needs of users, improve the user's service experience, and thus improve the work efficiency of call center staff.

[0005] Firstly, this application provides a product recommendation method, the method comprising: Obtain the browsing history of the products viewed by the user awaiting response, where the user awaiting response is the user waiting for a response from an agent. Based on the browsing history of the user to be responded to, determine the user intent characteristics of the user to be responded to, the user's interaction intent with the product, and the product characteristics of the product; Based on the interaction intent of the user to be responded to and the product characteristics of the product interacted with by the user to be responded to, the first fusion feature corresponding to the user to be responded to is determined; Based on the user's interaction intent with the product, the user intent characteristics, and the product characteristics, determine the second fusion feature corresponding to the product; Based on the first fusion feature corresponding to the user to be responded to and the second fusion feature corresponding to the product, determine whether the product is the target recommended product.

[0006] Secondly, this application also provides a product recommendation device, the device comprising: The acquisition module is used to acquire the browsing history of the products browsed by the user to be answered, wherein the user to be answered is the user waiting for the agent to respond; The extraction module is used to determine the user intent characteristics of the user to be responded to, the user's interaction intent with the product, and the product characteristics of the product based on the browsing history of the user to be responded to. The first fusion module is used to determine the first fusion feature corresponding to the user to be responded to based on the interaction intent of the user to be responded to and the product features of the product interacted with by the user to be responded to. The second fusion module is used to determine the second fusion feature corresponding to the product based on the user's interaction intent with the product, the user intent features, and the product features of the product. The target determination module is used to determine whether the product is a target recommended product based on the first fusion feature corresponding to the user to be responded to and the second fusion feature corresponding to the product.

[0007] Thirdly, this application also provides a computer device, which includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the product recommendation method as described above.

[0008] Fourthly, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the product recommendation method as described above.

[0009] This application provides a product recommendation method, apparatus, device, and storage medium. The method includes: acquiring the browsing history of a user awaiting response to browse products, wherein the user awaiting response is a user waiting for an agent's response; determining, based on the browsing history of the user awaiting response, the user's user intent characteristics, the user's interaction intent with the product, and the product characteristics of the product; determining, based on the user's interaction intent and the product characteristics of the product interacting with the user, determining a first fusion feature corresponding to the user awaiting response; determining, based on the user's interaction intent with the product, the user intent characteristics, and the product characteristics of the product, determining a second fusion feature corresponding to the product; and determining, based on the first fusion feature corresponding to the user and the second fusion feature corresponding to the product, whether the product is a target recommended product. This application, by determining the first fusion feature corresponding to the product interacting with by the user awaiting response and the second fusion feature corresponding to the user intent characteristics associated with the product interacting with the user, can analyze the impact of the product on the user awaiting response from two different perspectives: the impact of the product on the user awaiting response and the impact of the user awaiting response on the product. Then, by determining the target recommended product based on the first and second fusion features, it accurately recommends products that meet the needs of the user awaiting response, improving the service experience of the user awaiting response and thus improving the work efficiency of the agent. Taking the healthcare sector as an example, based on the fusion characteristics of user interactions regarding medical products such as recuperation services, health check-up packages, and nursing services, as well as the fusion characteristics of user intent features such as consultation and comparison associated with health check-up packages, it is possible to accurately determine whether a health check-up package meets the needs of the user. Similarly, in the financial sector, based on the fusion characteristics of user interactions regarding insurance products such as medical insurance, car insurance, accident insurance, and home insurance, as well as the fusion characteristics of user intent features associated with medical insurance, it is possible to accurately determine whether medical insurance meets the needs of the user. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating a product recommendation method provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the connection between the server and the terminal device provided in an embodiment of this application; Figure 3 A schematic diagram of a product recommendation scenario provided in an embodiment of this application; Figure 4 A schematic block diagram of a product recommendation device provided in an embodiment of this application; Figure 5 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0014] This application provides a product recommendation method, apparatus, device, and storage medium. The product recommendation method can be applied to terminal devices, such as mobile phones, tablets, laptops, and desktop computers. It can also be applied to servers, which can be standalone servers or cloud servers providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0015] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0016] Please see Figure 1 , Figure 1 This is a flowchart illustrating a product recommendation method provided in an embodiment of this application. It should be noted that the product recommendation method provided in this embodiment can be used on a terminal device, and of course, it can also be used on a server.

[0017] like Figure 2 As shown, the product recommendation method is applied to the server. The server and terminal devices communicate with each other, and the server can send the target recommended products obtained through the product recommendation method to the terminal devices. Of course, it is not limited to this, and no restrictions are imposed here.

[0018] In practice, terminal devices include, but are not limited to, any of the following: mobile phones, tablets, laptops, and desktop computers; servers can be standalone servers, server clusters, or cloud servers that provide cloud computing services.

[0019] like Figure 1 As shown, the recommended method for this product includes steps S101 to S105.

[0020] Step S101: Obtain the browsing history of the products viewed by the user waiting to be answered. The user waiting to be answered is the user waiting for the agent to respond.

[0021] Understandably, it's possible to access a user's browsing history before they successfully receive a response from an agent. For example, in the healthcare sector, this could involve accessing browsing history of different health checkup packages (elderly, men's, women's, and children's) on various websites, platforms, software, and mini-programs operated by medical checkup institutions. Similarly, in the financial sector, it could involve accessing browsing history of different car insurance products (compulsory traffic accident liability insurance, third-party liability insurance, vehicle damage insurance, etc.) on various websites, platforms, software, and mini-programs operated by insurance companies.

[0022] Step S102: Based on the browsing history of the user to be responded to, determine the user intent characteristics of the user to be responded to, the user's interaction intent with the product, and the product characteristics.

[0023] Based on the browsing history of the user awaiting a response to different car insurance products, such as viewing car insurance product details pages, comparing the prices or coverage of different car insurance products, and inquiring about prices, this application embodiment can determine that the user awaiting a response has the characteristics of a user who is initially seeking information, making comparisons and evaluations, or inquiring about prices.

[0024] Based on the user intent characteristics of the user awaiting a response, their interaction intent with the product can be determined. For example, based on the user's preliminary understanding of different car insurance products, it can be determined that the user does not have a clear purchase intention. However, based on the user's comparison of the prices of different car insurance products, it can be determined that the user is looking for the car insurance product with the best cost performance and has a clear purchase intention.

[0025] Based on the browsing history of different car insurance products viewed by the user awaiting a response, the product characteristics corresponding to each product can be obtained. For example, the compensation amount and insurance premium for compulsory traffic accident liability insurance applicable to non-commercial passenger cars, the insurance premium and coverage scope of vehicle damage insurance, etc.

[0026] Step S103: Determine the first fusion feature corresponding to the user to be responded to based on the interaction intent of the user to be responded to and the product characteristics of the product interacted with by the user to be responded to.

[0027] Understandably, the CF-PoolFormer attention mechanism can be used to fuse multiple interaction intentions of the user to be responded to with the product features of all products interacted with by the user. This allows for analysis and processing from the perspective of the impact of different products on the user, resulting in the first fused feature corresponding to the user. It should be noted that the CF-PoolFormer attention mechanism is a novel attention mechanism combining channel attention and frequency analysis. It can reduce the computational complexity in the feature fusion process, making it suitable for large-scale recommendation systems requiring real-time processing, such as the medical examination system of a medical examination company in the healthcare field and the insurance system of an insurance company in the financial field.

[0028] Specifically, during the browsing process, users typically interact with a large number of products. In this embodiment, the CF-PoolFormer attention mechanism can be used to first fuse multiple product features of each interacting product, and then fuse the product features corresponding to each of the multiple products. This reduces the number of product features aggregated during aggregation, decreases the amount of data to be calculated, and thus improves the efficiency of product recommendation.

[0029] Step S104: Determine the second fusion feature corresponding to the product based on the user's interaction intent with the product, the user intent characteristics, and the product characteristics.

[0030] It is understandable that each interaction intent of the user to be responded to regarding the product has multiple corresponding user intent features. This application embodiment can, based on the CF-PoolFormer attention mechanism, fuse the user's interaction intent with the product and the corresponding user intent features according to the product's characteristics, to analyze and process from the perspective of the user's impact on the product, thereby obtaining the second fused feature corresponding to the product.

[0031] For example, the user awaiting a response may have different interaction intentions regarding vehicle damage insurance, such as assessment, purchase, and after-sales service. Among these, when the interaction intention is to assess vehicle damage insurance, the corresponding user intention characteristics may include multiple user intention characteristics such as comparing premium prices and comparing coverage.

[0032] Specifically, the embodiments of this application can be based on the CF-PoolFormer attention mechanism, firstly fusing multiple user intent features corresponding to each interaction intent of the user to be responded to for the product, and then fusing the user intent features corresponding to each of the multiple interaction intents, so as to reduce the number of user intent features aggregated during aggregation, reduce the amount of data to be calculated, and thus improve the efficiency of product recommendation.

[0033] Step S105: Determine whether the product is the target recommended product based on the first fusion feature corresponding to the user to be responded to and the second fusion feature corresponding to the product.

[0034] It is understood that the first fusion feature in this application embodiment is the product feature of each product interacted with by the user to be responded to, and the second fusion feature is the user intent feature corresponding to the product interacted with by the user to be responded to. This application embodiment determines the recommendation level of the product by combining the perspective of the impact of different products on the user to be responded to and the perspective of the impact of the user to be responded to on the product, so as to determine whether the product meets the needs of the user to be responded to, and thus can determine whether the product is the target recommended product.

[0035] The product recommendation method provided in the above embodiments includes: obtaining the browsing history of a user waiting to be responded to, where the user is a user waiting for an agent's response; determining the user intent characteristics, the user's interaction intent with the product, and the product characteristics based on the user's browsing history; determining a first fusion characteristic corresponding to the user based on the user's interaction intent and the product characteristics of the product interacted with by the user; determining a second fusion characteristic corresponding to the product based on the user's interaction intent, user intent characteristics, and product characteristics; and determining whether the product is a target recommended product based on the first fusion characteristic corresponding to the user and the second fusion characteristic corresponding to the product. This embodiment of the application, by determining the first fusion characteristic corresponding to the product interacted with by the user and the second fusion characteristic corresponding to the user intent characteristics associated with the product interacted with by the user, can analyze the impact of the product on the user waiting to be responded to and the impact of the user waiting to be responded to on the product from two different perspectives: the impact of the product on the user waiting to be responded to and the impact of the user waiting to be responded to on the product. Then, based on the first and second fusion characteristics, a target recommended product is determined, accurately recommending products that meet the needs of the user waiting to be responded to, improving the service experience of the user waiting to be responded to, and thus improving the work efficiency of the agent. Taking the healthcare sector as an example, based on the fusion characteristics of user interactions regarding medical products such as recuperation services, health check-up packages, and nursing services, as well as the fusion characteristics of user intent features such as consultation and comparison associated with health check-up packages, it is possible to accurately determine whether a health check-up package meets the needs of the user. Similarly, in the financial sector, based on the fusion characteristics of user interactions regarding insurance products such as medical insurance, car insurance, accident insurance, and home insurance, as well as the fusion characteristics of user intent features associated with medical insurance, it is possible to accurately determine whether medical insurance meets the needs of the user.

[0036] In one exemplary embodiment, step S102 includes steps S1021 and S1022.

[0037] Step S1021: Based on the preset feature extraction network, extract the user intent features of the user to be responded to and the product features of the product according to the browsing history of the user to be responded to.

[0038] Step S1022: Based on the preset first transformation matrix, determine the user's interaction intent with the product according to the user intent characteristics of the user to be responded to.

[0039] For example, after obtaining the browsing history of the user to be responded to, the user intent features and product features of the user to be responded to can be extracted based on a preset feature extraction network. This application embodiment can divide the extraction process of user intent features and product features into two parts: one part is a feature extraction network for extracting user intent features, which may include a convolutional neural network or a graph neural network; the other part is a feature extraction network for extracting product features, which may include a pre-trained language model, a convolutional neural network, or a graph neural network. This application embodiment can also extract the user intent features of the user to be responded to and the product features of the product based on a dual-tower model.

[0040] This application embodiment, after obtaining the user intent characteristics of the user to be responded to and the product characteristics of the product, can determine the user's interaction intent with the product based on a preset first transformation matrix and the user intent characteristics of the user to be responded to. Specifically, a transformation matrix for extracting interaction intent can be designed for the user to be responded to, so as to obtain multiple interaction intents of the user to be responded to in interacting with the product. The first transformation matrix is: ; in, For the m-th interaction intent, p u For users awaiting response u User intent characteristics W m It is the transformation matrix of the m-th interaction intent.

[0041] In one exemplary embodiment, step S103 includes steps S1031 to S1033.

[0042] Step S1031: Based on the preset second transformation matrix, determine the product characteristics corresponding to each of the multiple preset interaction intentions according to the product characteristics of the product to be responded to by the user interaction.

[0043] Step S1032: Determine the product features corresponding to the multiple interaction intentions of the user to be responded to, based on the interaction intentions of the user to be responded to and the product features corresponding to the multiple preset interaction intentions.

[0044] Step S1033: Perform fusion processing on the product features corresponding to the multiple interaction intentions of the user to be responded to, and obtain the first fused feature corresponding to the user to be responded to.

[0045] For example, after obtaining the product characteristics of the product to be responded to by the user interaction in this application embodiment, the product characteristics corresponding to multiple preset interaction intentions can be determined based on the product characteristics of the product to be responded to by the user interaction, according to the preset second transformation matrix. This application embodiment can design a transformation matrix for extracting preset interaction intentions for the product, and the second transformation matrix is: ; in, For the product feature corresponding to the m-th preset interaction intent, q i For products i Product features, V m Let be the transformation matrix for the m-th preset interactive intent.

[0046] This application embodiment can be based on the CF-PoolFormer attention mechanism. According to the interaction intent of the user to be responded to and the product features corresponding to multiple preset interaction intents, the product features of the products interacting with the user under each interaction intent are fused to obtain the product features corresponding to each of the user's multiple interaction intents. Specifically, the probability that the user u interacts with product i due to the m-th interaction intent can be expressed as: ; att node The CF-PoolFormer attention mechanism can be specifically represented as follows: ; ; ; ; in, fea1 For user characteristics, fea2_reps For the repeated or extended form of product features that interact with users, g 1. To combine user characteristics and product characteristics, σ For activation function, W 2 is the weight matrix, and b2 is the bias term. g 2 is for g 1. Output results after linear transformation and nonlinear activation. This is the transpose of the weight matrix. g 3 is g2. The output result after normalization and pooling operations, plus itself.

[0047] The probability that user u, whose interaction intent is m, will interact with product i is obtained. Then, the weight coefficients are obtained by deregulation using the softmax function. : ; Where σ() is the activation function, exp() is the exponential function, and || is the join operation. Let m be the attention vector for the m-th interaction intent. For the m-th interaction intent of user u to be responded to, Let I be the product feature corresponding to the m-th preset interaction intent of product i. u All products that require a response from user interaction.

[0048] Furthermore, by integrating multiple product features corresponding to each user's interaction intent, the product features corresponding to each interaction intent can be obtained. For example, the product feature corresponding to the m-th interaction intent can be represented as: This yields the product characteristics corresponding to each of the multiple interactive intents of the user to be responded to. These product characteristics can be represented as follows: .

[0049] Finally, this embodiment of the application can also be based on the CF-PoolFormer attention mechanism to fuse the product features corresponding to the multiple interaction intents of the user to be responded to, thereby obtaining the first fused feature corresponding to the user to be responded to. Specifically, the product features corresponding to the multiple interaction intents of the user to be responded to are used as input, and the weight of each product feature is determined. The weight of each product feature can be expressed as follows: ; att int For the CF-PoolFormer attention mechanism with intent level, The weights of each product feature interacting with the user u to be responded to. The product features corresponding to each interaction intent of the user u to be responded to.

[0050] Considering that the weight of each product feature depends on the user u to be responded to, the product features and interaction intentions corresponding to each interaction intention of the user u to be responded to can be combined. For example, the product feature corresponding to the m-th interaction intention of the user u to be responded to. and the m-th interaction intent Combine them to learn their unified embedding , It can be represented as: ; in, C m This is the weight matrix. b m Let σ be the bias vector, σ be the activation function, and ‖ be the connection operation.

[0051] At this point, the weights of the product features corresponding to each interaction intent can be obtained, where the weight of the product feature corresponding to the m-th interaction intent is... w m It can be represented as: ; σ is the activation function. q T Here, b is the attention vector at the intent level, and b is the bias. Product features corresponding to the m-th interaction intent of user u to be responded to and the m-th interaction intent Embedded.

[0052] The weights of each product feature are then determined using the softmax function. Therefore, the weight of the product feature corresponding to the m-th interaction intent of the user u to be responded to can be specifically expressed as: ; Finally, the first fused feature is obtained by fusing the product features corresponding to multiple interaction intents. z u for: ; in, Let m be the weight of the product feature corresponding to the m-th interaction intent of the user u to be responded to. Let m be the product feature corresponding to user u's m-th interaction intent.

[0053] In one exemplary embodiment, step S104 includes steps S1041 to S1043.

[0054] Step S1041: Based on the preset second transformation matrix, determine the product features corresponding to each of the multiple preset interactive intentions according to the product features of the product.

[0055] Step S1042: Based on the user's interaction intent with the product, user intent characteristics, and product characteristics corresponding to each of the multiple preset interaction intents, determine the user intent characteristics corresponding to each of the multiple interaction intents of the user to be responded to.

[0056] Step S1043: Perform fusion processing on the user intent features corresponding to the multiple interaction intents of the user to be responded to regarding the product, and obtain the second fusion feature corresponding to the product.

[0057] For example, based on a preset second transformation matrix, the product characteristics corresponding to multiple preset interaction intentions can be determined according to the product characteristics of the product to be responded to by the user interaction. This application embodiment can design a transformation matrix for extracting preset interaction intentions for the product; the second transformation matrix is: ; in, For the product feature corresponding to the m-th preset interaction intent, q i For products i Product features, V m Let be the transformation matrix for the m-th preset interactive intent.

[0058] This application embodiment can be based on the CF-PoolFormer attention mechanism, which fuses multiple interaction intentions of the user to be responded to with the corresponding user intention features according to the product features corresponding to each of the multiple preset interaction intentions, to obtain the user intention features corresponding to each of the multiple interaction intentions of the user to be responded to with the product. Specifically, the probability that product i interacts with user u due to the m-th interaction intention can be expressed as: ; att node The CF-PoolFormer attention mechanism can be specifically represented as follows: ; ; ; ; in, fea1 As a product feature, fea2 For the repeated or expanded form of user characteristics that interact with the product, g 1. To combine product features and user features, σ For activation function, W 2 is the weight matrix. b 2 is the bias term. g 2 is for g 1. Output results after linear transformation and nonlinear activation. g 3 is g 2. The output result after normalization and pooling operations, plus itself. This is the transpose of the weight matrix.

[0059] The probability that product i interacts with user u due to the m-th interaction intent is obtained. Then, the weight coefficients are obtained by deregulation using the softmax function. : ; Where σ() is the activation function, exp() is the exponential function, and || is the join operation. Let m be the attention vector for the m-th interaction intent. For the m-th interaction intent of user u to be responded to, Let I be the product feature corresponding to the m-th preset interaction intent of product i. u All products that require a response from user interaction.

[0060] Furthermore, by aggregating multiple user intent features corresponding to the m-th interaction intent, we can obtain the user intent features corresponding to each interaction intent of the user to be responded to regarding product i. For example, the user intent feature corresponding to the m-th interaction intent of product i can be represented as: This yields the user intent features corresponding to each of the multiple interaction intents of the user to be responded to regarding product i. For example, the user intent features corresponding to each of the multiple interaction intents of the user to be responded to regarding product i can be represented as follows: .

[0061] Finally, this application embodiment can also be based on the CF-PoolFormer attention mechanism to fuse the user intent features corresponding to each of the multiple interaction intents of the user to be responded to regarding the product, thereby obtaining a second fused feature corresponding to the product. Specifically, the user intent features corresponding to each of the multiple interaction intents of the user to be responded to regarding the product are used as input, and the weight of each user intent feature is determined. The weight of each user intent feature can be expressed as: ; att int For the CF-PoolFormer attention mechanism with intent level, Let i be the weight of each user intent feature corresponding to product i. These are the user intent features for each interaction intent corresponding to product i.

[0062] Considering that the weight of each user intent feature depends on the product, the user intent features of product i corresponding to each interaction intent can be combined with the product features corresponding to the preset interaction intents. For example, the user intent features of product i corresponding to the m-th interaction intent can be combined. And the product features corresponding to the m-th preset interaction intent of product i. Combine them to learn their unified embedding , It can be represented as: ; in, C m This is the weight matrix. b m Let σ be the bias vector, σ be the activation function, and ‖ be the connection operation.

[0063] At this point, the weights of the user intent features corresponding to each interaction intent... w m It can be represented as: ; σ is the activation function. q T Here, b is the attention vector at the intent level, and b is the bias. The user intent feature corresponding to the m-th interaction intent of product i And the product features corresponding to the m-th preset interaction intent of product i. Embedded.

[0064] The weights of each user intent feature are then determined using the softmax function. Therefore, the weight of the user intent feature corresponding to the m-th interaction intent of product i can be specifically expressed as: ; Finally, the second fused feature is obtained by fusing the user intent features corresponding to multiple interaction intents. v i for: ; in, Let be the weight of the user intent feature corresponding to the m-th interaction intent of product i. Let m be the user intent feature corresponding to the m-th interaction intent of product i.

[0065] In one exemplary embodiment, step S105 includes steps S1051 and S1052.

[0066] Step S1051: Based on the preset prediction model, predict the recommendation score of the product according to the first fusion feature corresponding to the user to be responded to and the second fusion feature corresponding to the product.

[0067] Step S1052: When the recommended score is greater than or equal to a preset score threshold, determine the product corresponding to the recommended score as the target recommended product.

[0068] In one embodiment, the prediction model of this application may include a dual-tower model or a multilayer perceptron. Specifically, by inputting the first fused feature corresponding to the user to be responded to and the second fused feature corresponding to the product into the dual-tower model or the multilayer perceptron, the recommendation score of the product can be predicted.

[0069] In another embodiment, the prediction model of this application may include the CF-PoolFormer attention mechanism. Based on the CF-PoolFormer attention mechanism, a product recommendation score can be calculated according to the first fusion feature corresponding to the user to be responded to and the second fusion feature corresponding to the product. Specifically, the calculation process of the product recommendation score is as follows: ; ; ; ; in, z u The first fusion feature, v i This is the second fusion feature. g 1. To concatenate the first fusion feature and the second fusion feature, σ For activation function, W 2 is the weight matrix. b 2 is the bias term. g 2 is for g 1. Output results after linear transformation and nonlinear activation. g 3 is g 2. The output result after normalization and pooling operations, added to itself. This is the transpose of the weight matrix. This represents the recommendation score for product i for user u who is to receive a response.

[0070] It is understood that the objective function of the prediction model in the embodiments of this application is... L r It can be: ; in, O To predict the rating set, The predicted recommendation score for product i for user u who is to be responded to. r ui This represents the actual recommendation score of product i for user u who is waiting to respond.

[0071] This application embodiment can utilize a sparse regularizer to sparsify the first and second transformation matrices, thereby mitigating overparameterization and avoiding overfitting. Therefore, the adjusted objective function can be: ; Where θ represents the model parameters, θ = {W, V}, W is the first transformation matrix, V is the second transformation matrix, θ0 is the regularization term (e.g., sparsity metric), and λ is the hyperparameter. L r It is used to measure the difference between the model's predicted recommendation score and the actual recommendation score, and is reflected by a scoring method.

[0072] After obtaining the recommended scores for each product, the recommended scores of each product can be compared with the preset score thresholds, and products with recommended scores greater than or equal to the preset score thresholds can be selected as target recommended products.

[0073] This application embodiment determines the product recommendation score by combining the product's impact on the user awaiting response and the user's impact on the product, thereby improving the accuracy of product recommendations and increasing the efficiency of call center staff in handling users awaiting response.

[0074] In practical applications, such as Figure 3 As shown, user ratings for different products can be modeled as a bipartite graph g={U,I,R,ε} to determine the product recommendation score for each user. Here, U is a subset of N. u A set of N users, where I is N i Let R be a set of products, and R be the user ratings of the products. For each edge e = (u, i, r) ∈ ε, it represents a visible interaction, such as user u rating product i. The feature matrix of U is... , where L u User characteristics; the product feature matrix is ​​as follows: L i These are product features.

[0075] This application embodiment uses a bipartite graph of user interaction products and collaborative filtering technology based on the CF-PoolFormer attention mechanism to perform product recommendations. This can capture the complex relationship between users waiting to respond and products, thereby improving the accuracy of product recommendations.

[0076] In one exemplary embodiment, the method further includes steps S201 and S202.

[0077] Step S201: Display the product information of the target recommended product on the interface of the online chat session between the agent and the user waiting to be answered.

[0078] Step S202: In response to the agent's sending operation, the product information of the target recommended product is sent to the terminal device of the user waiting to be responded to.

[0079] Understandably, when agents engage in online chat sessions with users awaiting responses through an intelligent response system, the product information of the target recommended product can be displayed on the chat interface. This allows agents to quickly identify the target recommended product that meets the needs of the user awaiting response. In response to the agent's sending action, the product information of the target recommended product is sent to the user's terminal device, eliminating the need for agents to search manually and improving their work efficiency.

[0080] For example, on the interface of an online chat session between an agent and a user waiting to be answered, the product information of the target recommended product with the highest recommendation score can be displayed, or the product information of the target recommended products with a high recommendation score and no more than three can be displayed. This will not affect the visual effect of the interface, and will also allow the agent to quickly identify the target recommended product.

[0081] In one exemplary embodiment, the method further includes steps S301 to S303.

[0082] Step S301: When the agent is having an online chat session with the user to be answered, obtain the current description data of the user to be recommended.

[0083] Step S302: Determine the matching degree between the current description data and multiple target recommended products.

[0084] Step S303: Based on the matching degree between the current description data and multiple target recommended products, filter the multiple target recommended products to obtain the filtered target recommended products. The matching degree between the filtered target recommended products and the current description data is greater than the matching degree between other target recommended products and the current description data.

[0085] In the financial sector, current descriptive data can include vehicle information of the user awaiting response, such as whether it is a non-commercial vehicle, trolley, mileage, and number of claims. Based on this information, agents can more accurately recommend suitable auto insurance products to the user.

[0086] Understandably, the target recommended products are determined based on the browsing history of the user awaiting a response from an agent. During the online chat between the agent and the user, the user typically raises more specific needs. At this point, the agent can obtain the user's current description data to filter out products that better meet the user's needs from a pool of target recommended products, thereby improving the accuracy of product recommendations.

[0087] Specifically, embodiments of this application can predict the matching degree between the current description data of the user to be answered and multiple target recommended products based on a large language model or a convolutional neural network model, and select the target recommended product with the highest matching degree with the current description data from multiple target recommended products based on the matching degree between the current description data and multiple target recommended products.

[0088] Please see Figure 4 , Figure 4 This is a schematic block diagram of a product recommendation device provided in an embodiment of this application. The product recommendation device can be configured in a server or terminal device to execute the aforementioned product recommendation method.

[0089] like Figure 4 As shown, the recommended device for this product includes: an acquisition module 110, an extraction module 120, a first fusion module 130, a second fusion module 140, and a target determination module 150.

[0090] The acquisition module 110 is used to acquire the browsing history of the products browsed by the user to be answered. The user to be answered is the user waiting for the agent to respond.

[0091] The extraction module 120 is used to determine the user intent characteristics, the user's interaction intent with the product, and the product characteristics of the user to be responded to based on the user's browsing history.

[0092] The first fusion module 130 is used to determine the first fusion feature corresponding to the user to be responded to based on the interaction intent of the user to be responded to and the product features of the product interacted with by the user to be responded to.

[0093] The second fusion module 140 is used to determine the second fusion feature corresponding to the product based on the user's interaction intent with the product, the user intent characteristics, and the product characteristics.

[0094] The target determination module 150 is used to determine whether a product is a target recommended product based on the first fusion feature corresponding to the user to be responded to and the second fusion feature corresponding to the product.

[0095] In one exemplary embodiment, the extraction module 120 includes a feature extraction submodule and an intent extraction submodule.

[0096] The feature extraction submodule is used to extract the user intent features of the user to be responded to and the product features of the product based on the browsing history of the user to be responded to, using a preset feature extraction network.

[0097] The intent extraction submodule is used to determine the user's interaction intent with the product based on the user intent characteristics of the user to be responded to, according to the preset first transformation matrix.

[0098] In one exemplary embodiment, the first fusion module 130 includes a first feature determination submodule, a second feature determination submodule, and a first processing submodule.

[0099] The first feature determination submodule is used to determine the product features corresponding to each of multiple preset interaction intents based on the product features of the product to be responded to, according to the preset second transformation matrix.

[0100] The second feature determination submodule is used to determine the product features corresponding to the multiple interaction intentions of the user to be responded to, based on the interaction intention of the user to be responded to and the product features corresponding to the multiple preset interaction intentions.

[0101] The first processing submodule is used to fuse the product features corresponding to the multiple interaction intents of the user to be responded to, and obtain the first fused feature corresponding to the user to be responded to.

[0102] In one exemplary embodiment, the second fusion module 140 includes a third feature determination submodule, a fourth feature determination submodule, and a second processing submodule.

[0103] The third feature determination submodule is used to determine the product features corresponding to each of the multiple preset interactive intents based on the product features of the product and the preset second transformation matrix.

[0104] The fourth feature determination submodule is used to determine the user intent features corresponding to the multiple interaction intents of the user to be responded to regarding the product, based on the user intent features, the product features corresponding to the multiple preset interaction intents, and the user intent features corresponding to each of the multiple interaction intents.

[0105] The second processing submodule is used to fuse the user intent features corresponding to the multiple interaction intents of the user to be responded to regarding the product, and obtain the second fused feature corresponding to the product.

[0106] In one exemplary embodiment, the target determination module 150 includes a prediction submodule and a target determination submodule.

[0107] The prediction submodule is used to predict the recommendation score of a product based on a preset prediction model, according to the first fusion feature corresponding to the user to be responded to and the second fusion feature corresponding to the product.

[0108] The target determination submodule is used to determine the product corresponding to the recommended score as the target recommended product when the recommended score is greater than or equal to a preset score threshold.

[0109] In one exemplary embodiment, the apparatus further includes a display submodule and a transmission submodule.

[0110] The display submodule is used to display product information of the target recommended product on the interface of the online chat session between the agent and the user waiting to be answered.

[0111] The sending submodule is used to respond to the sending operation of the agent and send the product information of the target recommended product to the terminal device of the user waiting to be responded to.

[0112] In one exemplary embodiment, the apparatus further includes: an acquisition submodule, a matching degree determination submodule, and a filtering submodule.

[0113] The Acquisition submodule is used to obtain the current description data of the user to be recommended when the agent is having an online chat session with the user to be answered.

[0114] The matching degree determination submodule is used to determine the matching degree between the current description data and multiple target recommended products.

[0115] The filtering submodule is used to filter multiple target recommended products based on the matching degree between the current description data and multiple target recommended products, and obtain the filtered target recommended products. The matching degree between the filtered target recommended products and the current description data is greater than the matching degree between other target recommended products and the current description data.

[0116] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the above-described apparatus and its modules and units can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0117] The method of this application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0118] For example, the above-described methods and apparatus can be implemented as a computer program that can run on a computer device.

[0119] Please see Figure 5 , Figure 5 This is a schematic block diagram illustrating the structure of a computer device provided in an embodiment of this application. The computer device may be a server or a terminal device.

[0120] like Figure 5 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a storage medium and internal memory.

[0121] The storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform any recommended product method.

[0122] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0123] Internal memory provides an environment for the execution of computer programs stored in storage media. When these computer programs are executed by a processor, the processor can perform any product recommendation method.

[0124] This network interface is used for network communication, such as sending assigned tasks.

[0125] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0126] It should be understood that a processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other convertible logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0127] In one embodiment, the processor is configured to execute a computer program and, when executing the computer program, may perform the following steps: Retrieve the browsing history of the products viewed by the users awaiting response; the users awaiting response are those waiting for an agent to answer their questions. Based on the browsing history of the user to be responded to, determine the user intent characteristics of the user to be responded to, the user's interaction intent with the product, and the product characteristics of the product. Based on the interaction intent of the user to be responded to and the product characteristics of the product interacted with by the user to be responded to, determine the first fusion feature corresponding to the user to be responded to; Based on the user's interaction intent with the product, the user intent characteristics, and the product characteristics, determine the corresponding second fusion feature of the product; Based on the first fusion feature corresponding to the user to be responded to and the second fusion feature corresponding to the product, determine whether the product is the target recommended product.

[0128] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of product recommendation described above can be referred to the corresponding process in the aforementioned product recommendation method embodiments, and will not be repeated here.

[0129] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the following steps: Retrieve the browsing history of the products viewed by the users awaiting response; the users awaiting response are those waiting for an agent to answer their questions. Based on the browsing history of the user to be responded to, determine the user intent characteristics of the user to be responded to, the user's interaction intent with the product, and the product characteristics of the product. Based on the interaction intent of the user to be responded to and the product characteristics of the product interacted with by the user to be responded to, determine the first fusion feature corresponding to the user to be responded to; Based on the user's interaction intent with the product, the user intent characteristics, and the product characteristics, determine the corresponding second fusion feature of the product; Based on the first fusion feature corresponding to the user to be responded to and the second fusion feature corresponding to the product, determine whether the product is the target recommended product.

[0130] The computer-readable storage medium can be an internal storage unit of the computer device described in the foregoing embodiments, such as a hard disk or memory of the computer device. Alternatively, it can be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., provided on the computer device.

[0131] It should be noted that the functions or steps that can be achieved by the computer-readable storage medium described above can be referred to the embodiments of the aforementioned product recommended method.

[0132] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0133] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0134] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A product recommendation method characterized by, The method comprises: obtaining a browsing record of a user to be answered browsing a product, the user to be answered being a user waiting for an agent to answer; determining, according to the browsing record of the user to be answered, a user intention feature of the user to be answered, an interaction intention of the user to be answered on the product, and a product feature of the product; determining, according to the interaction intention of the user to be answered and the product feature of the product interacted by the user to be answered, a first fusion feature corresponding to the user to be answered; determining, according to the interaction intention of the user to be answered on the product, the user intention feature, and the product feature of the product, a second fusion feature corresponding to the product; determining, according to the first fusion feature corresponding to the user to be answered and the second fusion feature corresponding to the product, whether the product is a target recommended product.

2. The product recommendation method according to claim 1, characterized in that, The determining, according to the browsing record of the user to be answered, of the user intention feature of the user to be answered, the interaction intention of the user to be answered on the product, and the product feature of the product comprises: extracting, based on a preset feature extraction network, the user intention feature of the user to be answered and the product feature of the product according to the browsing record of the user to be answered; determining, based on a preset first conversion matrix, the interaction intention of the user to be answered on the product according to the user intention feature of the user to be answered.

3. The product recommendation method according to claim 1, characterized in that, The determining, according to the interaction intention of the user to be answered and the product feature of the product interacted by the user to be answered, of the first fusion feature corresponding to the user to be answered comprises: determining, based on a preset second conversion matrix, a product feature corresponding to each of a plurality of preset interaction intentions according to the product feature of the product interacted by the user to be answered; determining, according to the interaction intention of the user to be answered and the product feature corresponding to each of the plurality of preset interaction intentions, a product feature corresponding to each of a plurality of interaction intentions of the user to be answered; performing fusion processing on the product feature corresponding to each of the plurality of interaction intentions of the user to be answered to obtain the first fusion feature corresponding to the user to be answered.

4. The product recommendation method according to claim 1, characterized in that, The determining, according to the interaction intention of the user to be answered on the product, the user intention feature, and the product feature of the product, of the second fusion feature corresponding to the product comprises: determining, based on a preset second conversion matrix, a product feature corresponding to each of a plurality of preset interaction intentions according to the product feature of the product; determining, according to the interaction intention of the user to be answered on the product, the user intention feature, and the product feature corresponding to each of the plurality of preset interaction intentions, a user intention feature corresponding to each of a plurality of interaction intentions of the user to be answered on the product; performing fusion processing on the user intention feature corresponding to each of the plurality of interaction intentions of the user to be answered on the product to obtain the second fusion feature corresponding to the product.

5. The product recommendation method according to claim 1, characterized in that, The determining, according to the first fusion feature corresponding to the user to be answered and the second fusion feature corresponding to the product, of whether the product is a target recommended product comprises: predict a recommendation score of the product based on a preset prediction model and according to the first fusion feature corresponding to the user to be responded and the second fusion feature corresponding to the product; determine the product corresponding to the recommendation score as a target recommendation product when the recommendation score is greater than or equal to a preset score threshold.

6. The product recommendation method according to any one of claims 1 to 5, characterized in that, After determining the product as the target recommendation product, the method further comprises: display product information of the target recommendation product on an interface of an online chat session between the agent and the user to be responded; send the product information of the target recommendation product to a terminal device of the user to be responded in response to a sending operation of the agent.

7. The product recommendation method according to any one of claims 1 to 5, characterized in that, The method further comprises: obtain current description data of the user to be recommended when the agent and the user to be responded are in the online chat session; determine matching degrees of the current description data with a plurality of target recommendation products respectively; screen the plurality of target recommendation products according to the matching degrees of the current description data with the plurality of target recommendation products respectively to obtain screened target recommendation products, and the matching degree of the screened target recommendation products with the current description data is greater than the matching degrees of other target recommendation products with the current description data.

8. A product recommendation device characterized by comprising: The device comprises: an obtaining module configured to obtain a browsing record of a product browsed by a user to be responded, the user to be responded being a user waiting for response by an agent; an extracting module configured to determine, according to the browsing record of the user to be responded, a user intention feature of the user to be responded, an interaction intention of the user to be responded to a product, and a product feature of the product; a first fusion module configured to determine a first fusion feature corresponding to the user to be responded according to the interaction intention of the user to be responded and the product feature of the product interacted by the user to be responded; a second fusion module configured to determine a second fusion feature corresponding to the product according to the interaction intention of the user to be responded to the product, the user intention feature, and the product feature of the product; a target determining module configured to determine whether the product is a target recommendation product according to the first fusion feature corresponding to the user to be responded and the second fusion feature corresponding to the product.

9. A computer device, comprising: The computer device comprises a memory and a processor; the memory is configured to store a computer program; the processor is configured to execute the computer program and implement the product recommendation method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the product recommendation method according to any one of claims 1 to 7.

Images