Product recommendation method and device, storage medium and computer equipment

Abstract

The invention relates to the field of Internet technology and finance, and particularly discloses a product recommendation method and device, a storage medium and computer equipment, and the method comprises the steps: obtaining sample data; based on the sample data, training the initial recommendation model to obtain a target recommendation model; when any client accesses the dialogue system, obtaining second demand information and second user information of a target user corresponding to any client, and real-time product information and real-time risk information of each current product; and inputting the second demand information, the second user information, the real-time product information and the real-time risk information into a target recommendation model, obtaining a second product recommendation list corresponding to the target user based on model output of the target recommendation model, and returning the second product recommendation list to any client. According to the method, the user demand and the time-varying characteristics of financial products and the like are integrated in the product recommendation process, so that the accuracy and success rate of product recommendation can be greatly improved.

Description

Technical Field

[0001] The present application relates to the fields of Internet technology and finance, and in particular to a product recommendation method and device, a storage medium, and a computer device. Background Art

[0002] At present, with the continuous development of financial management products, various financial management products emerge in an endless stream. Different financial management products correspond to different risks, returns, etc., and their respective advantages also vary, making it difficult for users to choose financial management products.

[0003] In the prior art, when recommending financial products to users, most recommendations are made based on the user's historical purchasing habits or historical search habits, as well as the product features of the financial products. However, this method considers only a single factor for recommendation, resulting in low accuracy in the recommendation of financial products, which in turn affects the user's satisfaction with the recommendation. Summary of the invention

[0004] In view of this, the present application provides a product recommendation method and device, storage medium, and computer equipment. When recommending financial products and other products to target users, not only the user information of the target users is considered, but also the real-time needs of the target users, as well as the real-time product information and real-time risk information of the financial products, etc., and the user needs and the time-varying characteristics of the financial products, etc. are incorporated into the product recommendation process, so that the factors considered in product recommendations are more comprehensive, which can greatly improve the accuracy and success rate of product recommendations.

[0005] According to one aspect of the present application, a product recommendation method is provided, comprising:

[0006] Acquire sample data, where the sample data includes first demand information and first user information of a sample user, and historical product information and historical risk information corresponding to the time of the first demand information;

[0007] Based on the sample data, an initial recommendation model is trained to obtain a target recommendation model;

[0008] When any client accesses the dialogue system, the second demand information and second user information of the target user corresponding to the any client are obtained, as well as the real-time product information and real-time risk information of each current product in the business field corresponding to the dialogue system;

[0009] The second demand information, the second user information, the real-time product information and the real-time risk information are input into the target recommendation model, a second product recommendation list corresponding to the target user is obtained based on the model output of the target recommendation model, and the second product recommendation list is returned to any of the clients.

[0010] According to another aspect of the present application, a product recommendation device is provided, comprising:

[0011] A sample data acquisition module, used to acquire sample data, wherein the sample data includes first demand information and first user information of a sample user, and historical product information and historical risk information corresponding to the time of the first demand information;

[0012] A model training module, used to train the initial recommendation model based on the sample data to obtain a target recommendation model;

[0013] An information acquisition module, used for acquiring, when any client accesses the dialogue system, second demand information and second user information of a target user corresponding to the any client, and real-time product information and real-time risk information of current products in the business field corresponding to the dialogue system;

[0014] A list generation module is used to input the second demand information, the second user information, the real-time product information and the real-time risk information into the target recommendation model, obtain the second product recommendation list corresponding to the target user based on the model output of the target recommendation model, and return the second product recommendation list to any of the clients.

[0015] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned product recommendation method is implemented.

[0016] According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the above-mentioned product recommendation method when executing the program.

[0017] By means of the above technical solution, the present application provides a product recommendation method and device, storage medium, and computer equipment. First, sample data can be obtained. Here, the sample data can be composed of multiple parts, including the first demand information and first user information of the sample user. In addition, historical product information and historical risk information corresponding to the same time as the first demand information of the sample user can also be included. After obtaining the sample data, these sample data can be used to train the initial recommendation model, and the target recommendation model can be obtained after the training is completed. After obtaining the target recommendation model, the target recommendation model can be used to recommend products to the target user. Specifically, the second demand information, the second user information, the real-time product information, the real-time risk information, and the target recommendation model can be used to generate a second product recommendation list, and the second product recommendation list can be returned to the client. When recommending financial products and other products to target users, the embodiments of the present application not only consider the user information of the target users, but also consider the real-time needs of the target users, as well as the real-time product information and real-time risk information of the financial products and the like. The user needs and the time-varying characteristics of the financial products and the like are incorporated into the product recommendation process, so that the factors considered in product recommendations are more comprehensive, thereby greatly improving the accuracy and success rate of product recommendations.

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 A schematic diagram of a process flow of a product recommendation method provided in an embodiment of the present application is shown;

[0021] Figure 2 A schematic diagram showing a flow chart of another product recommendation method provided in an embodiment of the present application is shown;

[0022] Figure 3 A schematic diagram of a process flow of a product recommendation method provided in an embodiment of the present application is shown;

[0023] Figure 4 A schematic diagram of an initial recommendation model provided in an embodiment of the present application is shown;

[0024] Figure 5A schematic diagram of the structure of a product recommendation device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0025] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0026] In this embodiment, a product recommendation method is provided. Figure 1 As shown, the method includes:

[0027] Step 101, obtaining sample data, the sample data including first demand information and first user information of a sample user, and historical product information and historical risk information corresponding to the time of the first demand information;

[0028] The product recommendation method provided in the embodiment of the present application can be applied to the server side, and specifically can be applied to product recommendation scenarios such as financial management products. First, sample data can be obtained. Here, the sample data can be composed of multiple parts, including the first demand information and first user information of the sample user. In addition, it can also include historical product information and historical risk information corresponding to the same time as the first demand information of the sample user. Among them, the first demand information can be information about the specific needs of the sample user who wants to buy the product; the first user information can be information such as the sample user's occupation, age, gender, interests, and previously purchased products; the historical product information can be the release information, income information, transaction information, etc. of each product corresponding to the time when the sample user sends the first demand information; the historical risk information can be the trend information and forecast information of each product corresponding to the time when the sample user sends the first demand information. For example, when recommending financial management products, the sample users may be users who have purchased financial management products. The first demand information may be the specific demand of the sample users to purchase financial management products. The first user information may be the sample user's occupation, age, gender, interests, and previously purchased financial management products. Assuming that the sample user puts forward a demand to purchase financial management products on December 1, 2022, the historical product information may be the release information, income information, transaction information, etc. of each financial management product corresponding to December 1, 2022. The historical risk information may be the trend information and forecast information of each financial management product corresponding to December 1, 2022.

[0029] Step 102: training the initial recommendation model based on the sample data to obtain a target recommendation model;

[0030] In this embodiment, after the sample data are acquired, the initial recommendation model may be trained using the sample data, and a corresponding target recommendation model may be obtained after the training is completed.

[0031] Step 103, when any client accesses the dialogue system, obtain the second demand information and second user information of the target user corresponding to the any client, and the real-time product information and real-time risk information of each current product in the business field corresponding to the dialogue system;

[0032] In this embodiment, after obtaining the target recommendation model, the target recommendation model can be used to recommend products to the target user. In order to obtain the real-time needs of the target user, a dialogue system can be set up, and the target user can access the dialogue system through the client. Through the dialogue system, multiple rounds of interaction with the target user can be achieved, and then the needs of the target user are continuously matched in the process of dynamic interaction, so as to obtain the second demand information of the target user. Here, the dialogue system can be an intelligent voice dialogue system or a seat dialogue system. The second demand information can be information about the specific needs of the target user to buy the product. After detecting that the client of the target user is connected to the dialogue system, the second user information of the target user can also be directly obtained. The second user information can be information such as the occupation, age, gender, interests, and products purchased by the target user before. In addition, it can also be determined which business field the dialogue system accessed by the target user is a dialogue system. After determining the business field, the real-time product information and real-time risk information of all products in the business field can also be obtained. Among them, the real-time product information can be the release information, income information, transaction information, etc. of each product in the business field at the current time; the real-time risk information can be the trend information and forecast information of each product in the business field at the current time.

[0033] For example, in the financial management product recommendation scenario, if a target user wants to buy a financial management product but does not know which specific one to buy, he can call the financial management business service hotline. After calling, you can ask the target user which financial management product he wants to consult. Assuming that the target user wants to consult about fund management products, then at this time you can connect the target user's client to the dialogue system corresponding to the fund management product, and the dialogue system corresponding to the fund management product will inquire the target user. After multiple rounds of inquiries, the target user's second demand information is finally determined.

[0034] Step 104: input the second demand information, the second user information, the real-time product information and the real-time risk information into the target recommendation model, obtain the second product recommendation list corresponding to the target user based on the model output of the target recommendation model, and return the second product recommendation list to any one of the clients.

[0035] In this embodiment, after obtaining the second demand information, the second user information, the real-time product information and the real-time risk information, these information can be input into the target recommendation model, and the recommendation score of each product can be seen through the model output of the target recommendation model. Afterwards, a second product recommendation list for the target user can be generated based on these recommendation scores. Specifically, all products can be sorted according to the recommendation scores, and the second product recommendation list can be generated based on the top N products, or the second product recommendation list can be generated based on all sorted products.

[0036] By applying the technical solution of this embodiment, first, sample data can be obtained. Here, the sample data can be composed of multiple parts, including the first demand information and the first user information of the sample user. In addition, historical product information and historical risk information corresponding to the same time as the first demand information of the sample user can also be included. After obtaining the sample data, these sample data can be used to train the initial recommendation model, and the target recommendation model can be obtained after the training is completed. After obtaining the target recommendation model, the target recommendation model can be used to recommend products to the target user. Specifically, the second demand information, the second user information, the real-time product information, the real-time risk information and the target recommendation model can be used to generate a second product recommendation list, and the second product recommendation list is returned to the client. When recommending products such as financial management products to the target user, the embodiment of the present application not only considers the user information of the target user, but also considers the real-time needs of the target user, as well as the real-time product information and real-time risk information of financial management products, etc., and integrates the characteristics of user needs and financial management products that change over time in the product recommendation process, so that the factors considered when recommending products are more comprehensive, thereby greatly improving the accuracy and success rate of product recommendations.

[0037] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, another product recommendation method is provided, such as Figure 2 As shown, the method includes:

[0038] Step 201, obtaining sample data, the sample data including first demand information and first user information of a sample user, and historical product information and historical risk information corresponding to the time of the first demand information;

[0039] Step 202: input the sample data into the pre-trained graph neural network model to obtain a demand feature vector corresponding to the first demand information, a user feature vector corresponding to the first user information, a product feature vector corresponding to the historical product information, and a risk feature vector corresponding to the historical risk information;

[0040] In this embodiment, if Figure 3 and Figure 4 As shown, the initial recommendation model includes a pre-trained graph neural network model, an attention mechanism layer, and an MF model. Among them, the pre-trained graph neural network model can be a GPT-GNN model. In some cases, there are problems such as cold start, sparse data, and a small number of available labels in product recommendations. Therefore, pre-training GNN parameters in the form of GPT-GNN can enable downstream tasks to be trained with less labeled data. For example, for the recommendation scenario of financial management products, new registered users and new financial products have no relevant historical data, which makes the recommendation of financial management products have user cold start and product cold start problems, and the use of a pre-trained graph neural network model can better solve this problem. The present application can effectively reduce the demand for labeled data and reduce data annotation costs by pre-training the graph neural network model, and subsequently use it in downstream recommendation tasks, so that it can achieve better recommendation effects with less labeled data.

[0041] After obtaining the sample data, the sample data can be input into a pre-trained graph neural network model, which can then output a demand feature vector corresponding to the first demand information, a user feature vector corresponding to the first user information, a product feature vector corresponding to the historical product information, and a risk feature vector corresponding to the historical risk information.

[0042] Step 203, inputting the demand feature vector and the user feature vector into the first attention mechanism in the attention mechanism layer, performing feature vector fusion through the first attention mechanism, and obtaining a first fusion vector;

[0043] Step 204, inputting the product feature vector and the risk feature vector into the second attention mechanism in the attention mechanism layer, performing feature vector fusion through the second attention mechanism, and obtaining a second fusion vector;

[0044] In this embodiment, the initial recommendation model also includes an attention mechanism layer. After the demand feature vector, user feature vector, product feature vector and risk feature vector are output from the pre-trained graph neural network model, these vectors can also be input into the attention mechanism layer. Specifically, the attention mechanism layer can include two parts, one is the first attention mechanism and the other is the second attention mechanism. Among them, the first attention mechanism is used to input the demand feature vector and the user feature vector, fuse the demand feature vector and the user feature vector, and then output the first fusion vector; the second attention mechanism is used to input the product feature vector and the risk feature vector, fuse the product feature vector and the risk feature vector, and then output the second fusion vector.

[0045] Step 205: input the first fusion vector and the second fusion vector into the MF model, and obtain a first product recommendation list corresponding to the sample user based on a model output of the MF model;

[0046] In this embodiment, the initial recommendation model also includes an MF model. After obtaining the first fusion vector and the second fusion vector, the first fusion vector and the second fusion vector can be input into the MF (Full Name: Matrix Factorization; Chinese Name: Matrix Decomposition) model. After the operation of the MF model, the recommendation scores between the sample user and each product corresponding to the first demand information time can be output accordingly, and then the first product recommendation list corresponding to the sample user can be determined according to these recommendation scores.

[0047] Step 206, adjusting the model parameters in the initial recommendation model according to the first product recommendation list and the historical successful recommendation information of the sample user to obtain the target recommendation model;

[0048] In this embodiment, after obtaining the first product recommendation list corresponding to the sample user, the historical successful recommendation information corresponding to the sample user at the first demand information time can be obtained. The historical successful recommendation information may include products successfully purchased by the sample user, and the difference between the products successfully purchased by the sample user and the products included in the first product recommendation list is compared. The model loss value of the target recommendation model is determined based on the difference. Furthermore, the model parameters in the pre-trained graph neural network model, the attention mechanism layer, and the MF model in the initial recommendation model are adjusted through the model loss value, and finally the target recommendation model can be obtained.

[0049] Step 207, when any client accesses the dialogue system, obtain the second demand information and second user information of the target user corresponding to the any client, and the real-time product information and real-time risk information of each current product in the business field corresponding to the dialogue system;

[0050] Step 208: input the second demand information, the second user information, the real-time product information and the real-time risk information into the target recommendation model, obtain the second product recommendation list corresponding to the target user based on the model output of the target recommendation model, and return the second product recommendation list to any one of the clients.

[0051] In this embodiment, after obtaining the target recommendation model, the target recommendation model can be used to recommend products to the target user. Specifically, the second demand information, the second user information, the real-time product information, the real-time risk information and the target recommendation model can be used to generate a second product recommendation list, and the second product recommendation list can be returned to the client. Here, when the second user recommendation list is determined using the second demand information, the second user information, the real-time product information and the real-time risk information, similarly, the second demand information, the second user information, the real-time product information and the real-time risk information are input into the trained pre-trained graph neural network model of the target recommendation model to obtain the demand feature vector corresponding to the second demand information, the user feature vector corresponding to the second user information, the product feature vector corresponding to the real-time product information, and the risk feature vector corresponding to the real-time risk information. Then, the third fusion vector and the fourth fusion vector are obtained by using the trained first attention mechanism and the second attention mechanism, and the third fusion vector and the fourth fusion vector are input into the trained MF model. Based on the output of the trained MF model, the second product recommendation list is obtained. This process is consistent with the model training process of the target recommendation model.

[0052] In an embodiment of the present application, optionally, step 206 includes: obtaining a model loss value based on the products to be recommended in the first product recommendation list and the successfully recommended products indicated by the historical successful recommendation information; when the model loss value is greater than a preset loss threshold, adjusting the model parameters in the initial recommendation model, and returning to the step of inputting the sample data into the pre-trained graph neural network model until the model loss value is less than or equal to the preset loss threshold, thereby obtaining the target recommendation model.

[0053] In this embodiment, the first product recommendation list may include one or more products to be recommended. Then, the products to be recommended may be compared with the successfully recommended products indicated by the historical successful recommendation information, and the model loss value may be calculated by the model loss function. Specifically, there is no special requirement for the model loss function, as long as the model loss value can be calculated based on the products to be recommended and the successfully recommended products. If the model loss value is greater than the preset loss threshold, it means that the model accuracy of the initial recommendation model is not enough at this time, and the model parameters in the initial recommendation model may be adjusted to obtain an updated initial recommendation model. Afterwards, the sample data may be re-input into the updated initial recommendation model to obtain an updated first product recommendation list, and based on the updated first product recommendation list and the successfully recommended products indicated by the historical successful recommendation information, the model loss value may be calculated again until the model loss value is less than or equal to the preset loss threshold. When the model loss value is less than or equal to the preset loss threshold, it means that the model accuracy of the updated initial recommendation model has reached the requirement, and the updated initial recommendation model may be used as the target recommendation model.

[0054] In an embodiment of the present application, optionally, the "when any client accesses the dialogue system, obtaining the second demand information of the target user corresponding to the any client" in step 207 includes: when any client accesses the dialogue system, calling the corresponding inquiry model based on the business field corresponding to the dialogue system; according to the inquiry model, obtaining multiple rounds of interaction data corresponding to the target user, and extracting the sub-demand information of the target user from each round of interaction data; fusing the sub-demand information to obtain the second demand information of the target user.

[0055] In this embodiment, if a client is connected to the dialogue system, the business field corresponding to the dialogue system can be identified. For example, for the recommendation scenario of financial products, the business field may include gold business, savings business, fund business, stock business, bond business, insurance business, etc., and each business field may correspond to a dialogue system. Each dialogue system corresponds to an inquiry model, and multiple rounds of interaction with the target user can be realized through the inquiry model. The input of the inquiry model may be the inquiry text corresponding to the inquiry voice of the target user, and the output may be the corresponding reply text. Each round of interaction data may include the inquiry text and the reply text, and the sub-demand information of the target user can be obtained from each round of interaction data. Specifically, when the target user chooses to consult the fund business, the dialogue system corresponding to the fund business calls the inquiry model corresponding to the fund business, and can convert the inquiry voice of the target user into inquiry text, which is input into the fund business inquiry model accordingly. Then, the fund business inquiry model can query the reply text according to the inquiry text. Since the inquiry model is obtained by training multiple labeled data in advance, the inquiry intention of the target user can be determined more accurately according to the inquiry text, and the corresponding reply text can be matched. After obtaining the reply text, the reply text can be converted into voice and output to the client of the target user.

[0056] For example, when the target user chooses to consult fund business, the target user asks in the first round "What are the funds with high interest rates at present?", and the corresponding reply text is "a"; the target user asks in the second round "Which fund has lower risk?", and the corresponding reply text is "b"; the target user asks in the third round "Can you introduce product A of fund?", and the corresponding reply text is "c". Then the sub-demand information of the target user can be extracted from each of the three rounds of interaction data, for example, the sub-demand information extracted in the first round is "funds with high interest rates", the sub-demand information extracted in the second round is "funds with low risk", and the sub-demand information extracted in the third round is "Fund A".

[0057] After obtaining each sub-demand information, these sub-demand information can be merged together to obtain the second demand information of the target user. The embodiment of the present application obtains the real-time demand of the target user through the dialogue system, which can make the determination of user demand more accurate, which is conducive to improving the accuracy of subsequent product recommendations, and can also effectively alleviate the shortcomings of user active search and improve user satisfaction.

[0058] In an embodiment of the present application, optionally, the "obtaining the second user information of the target user corresponding to any one of the clients" in step 207 includes: obtaining basic information of the target user and historical product purchase information of the target user in the business field corresponding to the dialogue system, the basic information including at least one of the occupation, age, gender, and interests of the target user; and using the basic information and the historical product purchase information as the second user information of the target user.

[0059] In this embodiment, after the client of the target user is connected to the dialogue system, on the one hand, the basic information of the target user can be obtained, wherein the basic information is usually fixed information, such as the occupation, age, gender, usual interests and hobbies of the target user, etc., through which the basic situation of the target user can be understood. According to statistics, users of different occupations, different age groups, different genders, etc. have certain tendencies in purchasing financial products. Therefore, obtaining the basic information of the target user can improve the success rate of financial product recommendation to a certain extent. On the other hand, the historical product purchase information of the target user in the business field corresponding to the access dialogue system can also be obtained. For example, if the business field corresponding to the accessed dialogue system is the fund business field, then the relevant information of the fund previously purchased by the target user can be found, and this information can be used as the historical product purchase information of the target user. Through the historical product purchase information, the historical purchase situation of the target user in the business field can be known, and it can also be used as auxiliary information for product recommendation to a certain extent. Finally, the basic information and historical product purchase information of the target user can be combined to determine the second user information of the target user. The first user information of the above sample user can also be obtained in this way.

[0060] In an embodiment of the present application, optionally, the "obtaining real-time product information and real-time risk information of each current product in the business field corresponding to the dialogue system" in step 207 includes: determining the business field corresponding to the dialogue system, and all business products under the business field; obtaining real-time data corresponding to all business products, and extracting the real-time product information from the real-time data, the real-time product information including at least one of product release information, product revenue information and product transaction information; obtaining product trend information and product forecast information corresponding to all business products, and determining the real-time risk information based on the product trend information and the product forecast information.

[0061] In this embodiment, when determining the real-time product information and real-time risk information, first, the business field corresponding to the access dialogue system and all business products under the business field can be determined. Then, on the one hand, the real-time data of these business products can be obtained, and on the other hand, the product trend information and product forecast information corresponding to these business products can be obtained. Here, the real-time data can include the real-time product information of the current business products, such as product release information, product revenue information, product transaction information, etc. From the product release information, the release time and publisher of the business product can be determined, from the product revenue information, the change of the business product's revenue from the release to the present can be determined, and from the product transaction information, the current transaction of the business product can be determined. In addition, from the product trend information, the transaction volume and transaction price of the business product from the release to the present can be determined, and from the product forecast information, the future trend of the business product can be determined. Through the product trend information and product forecast information, the real-time risk information of the business product can be finally determined. The above-mentioned historical product information and historical risk information can also be obtained in this way.

[0062] For example, for the fund business field, all business products under the fund business field currently include Fund 1, Fund 2, Fund 3, Fund 4 and Fund 5. Then the real-time data, product trend information, and product forecast information of these five funds can be obtained, and finally the real-time product information of these five funds can be determined from the real-time data, and the real-time risk information of these five funds can be determined based on the product trend information and product forecast information.

[0063] When recommending products to target users, this application not only considers the target users' real-time demand information and user information, but also adds real-time product information and real-time risk information, making the success rate of product recommendations higher. Especially for the recommendation scenario of financial management products, considering the particularity of financial management products, that is, their own value will change over time and they are also easily affected by market risks, this application adds time factors and risk factors to product information, so as to better describe financial management products, which can greatly improve the recommendation effect.

[0064] In an embodiment of the present application, optionally, after step 208, the method further includes: receiving recommendation feedback information returned by any of the clients, and determining the recommendation satisfaction corresponding to the recommendation feedback information; when there are a preset number of recommendation feedback information whose recommendation satisfactions are all less than a preset satisfaction, updating the target recommendation model.

[0065] In this embodiment, after recommending products to the target user, the model parameters of the target recommendation model can be adjusted according to the recommendation feedback information of the target user. Specifically, if the accumulated preset number of recommendation feedback information all indicate that the user's recommendation satisfaction is less than the preset satisfaction, it means that the recommendation effect of the target recommendation model is still poor. At this time, the model parameters in the target recommendation model can be adjusted, thereby updating the target recommendation model.

[0066] Further, as Figure 1 The specific implementation of the method, the embodiment of the present application provides a product recommendation device, such as Figure 5 As shown, the device comprises:

[0067] A sample data acquisition module, used to acquire sample data, wherein the sample data includes first demand information and first user information of a sample user, and historical product information and historical risk information corresponding to the time of the first demand information;

[0068] A model training module, used to train the initial recommendation model based on the sample data to obtain a target recommendation model;

[0069] An information acquisition module, used for acquiring, when any client accesses the dialogue system, second demand information and second user information of a target user corresponding to the any client, and real-time product information and real-time risk information of current products in the business field corresponding to the dialogue system;

[0070] A list generation module is used to input the second demand information, the second user information, the real-time product information and the real-time risk information into the target recommendation model, obtain the second product recommendation list corresponding to the target user based on the model output of the target recommendation model, and return the second product recommendation list to any of the clients.

[0071] Optionally, the initial recommendation model includes a pre-trained graph neural network model, an attention mechanism layer and an MF model; the model training module includes:

[0072] a data input unit, configured to input the sample data into the pre-trained graph neural network model to obtain a demand feature vector corresponding to the first demand information, a user feature vector corresponding to the first user information, a product feature vector corresponding to the historical product information, and a risk feature vector corresponding to the historical risk information;

[0073] A first fusion unit, configured to input the demand feature vector and the user feature vector into a first attention mechanism in the attention mechanism layer, and perform feature vector fusion through the first attention mechanism to obtain a first fusion vector;

[0074] A second fusion unit is used to input the product feature vector and the risk feature vector into a second attention mechanism in the attention mechanism layer, and perform feature vector fusion through the second attention mechanism to obtain a second fusion vector;

[0075] a vector input unit, configured to input the first fusion vector and the second fusion vector into the MF model, and obtain a first product recommendation list corresponding to the sample user based on a model output of the MF model;

[0076] A parameter adjustment unit is used to adjust the model parameters in the initial recommendation model according to the first product recommendation list and the historical successful recommendation information of the sample user to obtain the target recommendation model.

[0077] Optionally, the parameter adjustment unit is further used to:

[0078] Based on the products to be recommended in the first product recommendation list and the successfully recommended products indicated by the historical successful recommendation information, a model loss value is obtained; when the model loss value is greater than a preset loss threshold, the model parameters in the initial recommendation model are adjusted, and the process returns to the step of inputting the sample data into the pre-trained graph neural network model until the model loss value is less than or equal to the preset loss threshold, thereby obtaining the target recommendation model.

[0079] Optionally, the information acquisition module is further used to:

[0080] When any client accesses the dialogue system, the corresponding inquiry model is called based on the business field corresponding to the dialogue system; based on the inquiry model, multiple rounds of interaction data corresponding to the target user are obtained, and the sub-demand information of the target user is extracted from each round of interaction data; the sub-demand information is fused to obtain the second demand information of the target user.

[0081] Optionally, the information acquisition module is further used to:

[0082] Obtain basic information of the target user and historical product purchase information of the target user in the business field corresponding to the dialogue system, wherein the basic information includes at least one of the occupation, age, gender, and interests of the target user; and use the basic information and the historical product purchase information as second user information of the target user.

[0083] Optionally, the information acquisition module is further used to:

[0084] Determine the business field corresponding to the dialogue system and all business products under the business field; obtain real-time data corresponding to all business products, and extract the real-time product information from the real-time data, wherein the real-time product information includes at least one of product release information, product revenue information and product transaction information; obtain product trend information and product forecast information corresponding to all business products, and determine the real-time risk information based on the product trend information and the product forecast information.

[0085] Optionally, the device further comprises:

[0086] a feedback information receiving module, configured to receive recommendation feedback information returned by any client after returning the second product recommendation list to any client, and determine the recommendation satisfaction level corresponding to the recommendation feedback information;

[0087] The model updating module is used to update the target recommendation model when there is a preset number of recommendation feedback information whose recommendation satisfaction levels are all less than a preset satisfaction level.

[0088] It should be noted that for other corresponding descriptions of the functional units involved in the product recommendation device provided in the embodiment of the present application, please refer to Figures 1 to 4 The corresponding description in the method will not be repeated here.

[0089] Based on the above Figures 1 to 4 The method shown in the embodiment of the present application accordingly provides a storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned Figures 1 to 4 Product recommendation method shown.

[0090] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0091] Based on the above Figures 1 to 4 The method shown, and Figure 5 In order to achieve the above-mentioned purpose, the embodiment of the present application further provides a computer device, which can be a personal computer, a server, a network device, etc. The computer device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figures 1 to 4 Product recommendation method shown.

[0092] Optionally, the computer device may further include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc.

[0093] Those skilled in the art will appreciate that the computer device structure provided in this embodiment does not limit the computer device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0094] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages and saves the hardware and software resources of the computer device, and supports the operation of information processing programs and other software and / or programs. The network communication module is used to realize communication between the components inside the storage medium, and communication with other hardware and software in the physical device.

[0095] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform, or by hardware. First, sample data can be obtained. Here, the sample data can be composed of multiple parts, including the first demand information and first user information of the sample user. In addition, historical product information and historical risk information corresponding to the same time as the first demand information of the sample user can also be included. After obtaining the sample data, these sample data can be used to train the initial recommendation model, and the target recommendation model can be obtained after the training is completed. After obtaining the target recommendation model, the target recommendation model can be used to recommend products to the target user. Specifically, the second demand information, the second user information, the real-time product information, the real-time risk information and the target recommendation model can be used to generate a second product recommendation list, and the second product recommendation list can be returned to the client. When recommending financial products and other products to target users, the embodiments of the present application not only consider the user information of the target users, but also consider the real-time needs of the target users, as well as the real-time product information and real-time risk information of the financial products and the like. The user needs and the time-varying characteristics of the financial products and the like are incorporated into the product recommendation process, so that the factors considered in product recommendations are more comprehensive, thereby greatly improving the accuracy and success rate of product recommendations.

[0096] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple submodules.

[0097] The above serial numbers of this application are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of this application, but this application is not limited to them, and any changes that can be thought of by technicians in this field should fall within the scope of protection of this application.

Claims

1. A product recommendation method, characterized in that: include: Acquire sample data, where the sample data includes first demand information and first user information of a sample user, and historical product information and historical risk information corresponding to the time of the first demand information; Based on the sample data, an initial recommendation model is trained to obtain a target recommendation model; When any client accesses the dialogue system, the second demand information and second user information of the target user corresponding to the any client are obtained, as well as the real-time product information and real-time risk information of each current product in the business field corresponding to the dialogue system; The second demand information, the second user information, the real-time product information and the real-time risk information are input into the target recommendation model, a second product recommendation list corresponding to the target user is obtained based on the model output of the target recommendation model, and the second product recommendation list is returned to any of the clients.

2. The method according to claim 1, characterized in that The initial recommendation model includes a pre-trained graph neural network model, an attention mechanism layer, and an MF model; the initial recommendation model is trained based on the sample data to obtain a target recommendation model, including: Inputting the sample data into the pre-trained graph neural network model to obtain a demand feature vector corresponding to the first demand information, a user feature vector corresponding to the first user information, a product feature vector corresponding to the historical product information, and a risk feature vector corresponding to the historical risk information; Inputting the demand feature vector and the user feature vector into a first attention mechanism in the attention mechanism layer, and fusing the feature vectors through the first attention mechanism to obtain a first fused vector; Inputting the product feature vector and the risk feature vector into a second attention mechanism in the attention mechanism layer, and fusing the feature vectors through the second attention mechanism to obtain a second fused vector; Inputting the first fusion vector and the second fusion vector into the MF model, and obtaining a first product recommendation list corresponding to the sample user based on a model output of the MF model; According to the first product recommendation list and the historical successful recommendation information of the sample user, the model parameters in the initial recommendation model are adjusted to obtain the target recommendation model.

3. The method according to claim 2, characterized in that The adjusting the model parameters in the initial recommendation model according to the first product recommendation list and the historical successful recommendation information of the sample user to obtain the target recommendation model includes: Obtaining a model loss value based on the products to be recommended in the first product recommendation list and the successfully recommended products indicated by the historical successful recommendation information; When the model loss value is greater than the preset loss threshold, the model parameters in the initial recommendation model are adjusted, and the process returns to the step of inputting the sample data into the pre-trained graph neural network model until the model loss value is less than or equal to the preset loss threshold, thereby obtaining the target recommendation model.

4. The method according to claim 1, characterized in that: When any client accesses the dialogue system, obtaining the second demand information of the target user corresponding to any client includes: When any client accesses the dialogue system, the corresponding query model is called based on the business domain corresponding to the dialogue system; According to the inquiry model, multiple rounds of interaction data corresponding to the target user are acquired, and sub-demand information of the target user is extracted from each round of interaction data; The sub-demand information is merged to obtain the second demand information of the target user.

5. The method according to claim 1, characterized in that The obtaining the second user information of the target user corresponding to any one of the clients includes: Acquire basic information of the target user and historical product purchase information of the target user in the business field corresponding to the dialogue system, wherein the basic information includes at least one of the occupation, age, gender, and interests of the target user; The basic information and the historical product purchase information are used as the second user information of the target user.

6. The method according to claim 1, characterized in that Acquire real-time product information and real-time risk information of each current product in the business field corresponding to the dialogue system, including: Determine the business domain corresponding to the dialogue system and all business products under the business domain; Acquire real-time data corresponding to all business products, and extract the real-time product information from the real-time data, wherein the real-time product information includes at least one of product release information, product revenue information, and product transaction information; Obtain product trend information and product forecast information corresponding to all the business products, and determine the real-time risk information based on the product trend information and the product forecast information.

7. The method according to any one of claims 1 to 6, characterized in that After returning the second product recommendation list to any one of the clients, the method further includes: receiving recommendation feedback information returned by any of the clients, and determining a recommendation satisfaction level corresponding to the recommendation feedback information; When there is a preset number of recommendation feedback information whose recommendation satisfaction levels are all less than a preset satisfaction level, the target recommendation model is updated.

8. A product recommendation device, characterized in that: include: A sample data acquisition module, used to acquire sample data, wherein the sample data includes first demand information and first user information of a sample user, and historical product information and historical risk information corresponding to the time of the first demand information; A model training module, used to train the initial recommendation model based on the sample data to obtain a target recommendation model; An information acquisition module, used for acquiring, when any client accesses the dialogue system, second demand information and second user information of a target user corresponding to the any client, and real-time product information and real-time risk information of current products in the business field corresponding to the dialogue system; A list generation module is used to input the second demand information, the second user information, the real-time product information and the real-time risk information into the target recommendation model, obtain the second product recommendation list corresponding to the target user based on the model output of the target recommendation model, and return the second product recommendation list to any one of the clients.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images