Product recommendation method and device, storage medium and computer equipment

Abstract

The invention discloses a product recommendation method and device, a storage medium and computer equipment, relates to the technical field of natural language processing and the field of financial science and technology, and mainly aims at solving the problem that existing product recommendation is low in accuracy. The method mainly comprises the following steps: obtaining subjective feature information of a target user, wherein the subjective feature information is extracted based on a consultation dialogue text of the target user; determining an initial product recommendation result based on the relevancy between the subjective feature information and the candidate product combination, and obtaining objective evaluation information of the initial product recommendation result; generating a target product recommendation result according to the objective evaluation information, the subjective feature information and the initial product recommendation result through a trained product recommendation model; and sending the target product recommendation result to a target user terminal. The method is mainly used for recommending financial products.

Description

Technical Field

[0001] The present invention relates to the fields of natural language processing technology and financial technology, and in particular to a product recommendation method and device, a storage medium, and a computer device. Background Art

[0002] With the rapid development of my country's financial technology services and the enhancement of national investment awareness, the purchase of investment products has become more and more professional. More investors choose to purchase investment products through the channels of investment service companies and receive more professional investment consulting guidance to achieve more scientific and reasonable investment product configuration.

[0003] The current investment product recommendations are mainly based on the customer's purchasing power and risk tolerance to select product combinations from pre-made product combinations, or recommend popular product combinations to customers based on the popularity of the products. However, this method only evaluates the product's recommendability to customers from the perspective of the product's own risks and target dimensions. The evaluation dimension is single, and the recommended results have a low match with the customer, resulting in low accuracy of product recommendations. Summary of the invention

[0004] In view of this, the present invention provides a product recommendation method and device, storage medium, and computer equipment, the main purpose of which is to solve the problem of low recommendation accuracy of existing products, especially financial investment products.

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

[0006] Obtaining subjective characteristic information of a target user, wherein the subjective characteristic information is extracted based on a consultation dialogue text of the target user;

[0007] Determine an initial product recommendation result based on the relevance between the subjective feature information and the candidate product combination, and obtain objective evaluation information of the initial product recommendation result;

[0008] Generate a target product recommendation result according to the objective evaluation information, the subjective feature information, and the initial product recommendation result through the trained product recommendation model;

[0009] The target product recommendation result is sent to the target user terminal.

[0010] Furthermore, determining the initial product recommendation result based on the correlation between the subjective feature information and the candidate product combination includes:

[0011] Using the trained relevance prediction model, predicting the candidate product combinations according to the subjective feature information, and obtaining the relevance of different candidate product combinations;

[0012] Sorting the candidate product combinations in descending order according to the relevance;

[0013] Generate initial product recommendation results based on candidate product combinations whose rankings meet a preset ranking threshold.

[0014] Furthermore, before the candidate products are predicted based on the consultation dialogue text by using the trained relevance prediction model to obtain the initial recommended product results, the method further includes:

[0015] Obtaining a first training sample set, and building an initial relevance prediction model and an initial product recommendation model;

[0016] Using the first training sample set to train the initial relevance prediction model to obtain a trained relevance prediction model;

[0017] Constructing a second training sample set based on the output result of the trained relevance prediction model, wherein the second training sample set includes objective evaluation information samples;

[0018] The initial product recommendation model is trained using the second training sample set to obtain a trained product recommendation model.

[0019] Furthermore, the initial product recommendation result includes at least one product combination to be recommended, and the objective evaluation information of obtaining the initial product recommendation result includes:

[0020] Acquire industry information matching the product combination to be recommended from a product industry information database;

[0021] Sending the subjective characteristic information, the initial product recommendation result, and the industry information to an evaluation terminal, so that the evaluation terminal evaluates the product combination to be recommended;

[0022] Receive objective evaluation information of the to-be-recommended product combination returned by the evaluation terminal, wherein the objective evaluation information includes at least one of a recommendation parameter and an evaluation reference factor.

[0023] Furthermore, before obtaining the subjective characteristic information of the target user, the method further includes:

[0024] Receive the inquiry question text of the target user;

[0025] Generate a reply text to the consulting question text according to the consulting question text by using the trained consulting reply model;

[0026] Sending the reply text to the terminal of the target user to obtain the next consultation question text of the target user;

[0027] In response to an end instruction indicating the end of the consultation with the target user, subjective feature information is generated according to the consultation question text through the trained consultation response model.

[0028] Furthermore, before obtaining the subjective characteristic information of the target user, the method further includes:

[0029] Acquire a consultation dialogue text of the target user, where the consultation dialogue text is generated based on a speech recognition model;

[0030] Feature information is extracted from the user-side dialogue text in the consultation dialogue text using a trained feature extraction model to obtain subjective feature information.

[0031] Furthermore, after the trained product recommendation model generates a target product recommendation result according to the objective evaluation information, the subjective feature information, and the initial product recommendation result, the method further includes:

[0032] If the candidate product combination does not include the target recommended product combination, the target recommended product combination is updated to the candidate product combination.

[0033] According to another aspect of the present invention, a product recommendation device is provided, comprising:

[0034] An acquisition module, used to acquire subjective characteristic information of a target user, wherein the subjective characteristic information is extracted based on a consultation dialogue text of the target user;

[0035] A determination module, configured to determine an initial product recommendation result based on the relevance between the subjective feature information and the candidate product combination, and obtain objective evaluation information of the initial product recommendation result;

[0036] A prediction module, configured to generate a target product recommendation result according to the objective evaluation information, the subjective feature information, and the initial product recommendation result through a trained product recommendation model;

[0037] The sending module is used to send the target product recommendation result to the target user terminal.

[0038] Furthermore, the determining module includes:

[0039] A prediction unit, configured to perform prediction processing on the candidate product combinations according to the subjective feature information by using a trained relevance prediction model, so as to obtain relevances of different candidate product combinations;

[0040] A sorting unit, used to sort the candidate product combinations in descending order according to the relevance;

[0041] The generating unit is used to generate an initial product recommendation result based on a combination of candidate products whose rankings meet a preset ranking threshold.

[0042] Furthermore, the device also includes:

[0043] A first construction module is used to obtain a first training sample set and construct an initial relevance prediction model and an initial product recommendation model;

[0044] A first training module, used to train the initial relevance prediction model using the first training sample set to obtain a trained relevance prediction model;

[0045] A second construction module is used to construct a second training sample set based on the output result of the trained relevance prediction model, wherein the second training sample set includes objective evaluation information samples;

[0046] The second training module is used to train the initial product recommendation model using the second training sample set to obtain a trained product recommendation model.

[0047] Furthermore, the determining module further includes:

[0048] An acquisition unit, used for acquiring industry information matching the product combination to be recommended from a product industry information database;

[0049] A sending unit, configured to send the subjective characteristic information, the initial product recommendation result, and the industry information to an evaluation terminal, so that the evaluation terminal evaluates the product combination to be recommended;

[0050] The first receiving unit is configured to receive objective evaluation information of the to-be-recommended product combination returned by the evaluation terminal, wherein the objective evaluation information includes at least one of a recommendation parameter and an evaluation reference factor.

[0051] Furthermore, the device also includes:

[0052] A receiving module, used for receiving the consulting question text of the target user;

[0053] A first generating module is used to generate a reply text of the consulting question text according to the consulting question text by using the trained consulting reply model;

[0054] The sending module is further used to send the reply text to the terminal of the target user to obtain the next consultation question text of the target user;

[0055] The second generating module is used to generate subjective feature information according to the consultation question text through the trained consultation response model in response to an end instruction indicating the end of the consultation of the target user.

[0056] Furthermore, the device also includes:

[0057] The acquisition module is further used to acquire the consultation dialogue text of the target user, where the consultation dialogue text is generated based on the speech recognition model;

[0058] The extraction module is used to extract feature information from the user-side dialogue text in the consultation dialogue text through a trained feature extraction model to obtain subjective feature information.

[0059] Furthermore, the device also includes:

[0060] An updating module is used for updating the target recommended product combination into the candidate product combination if the candidate product combination does not include the target recommended product combination.

[0061] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to execute operations corresponding to the above-mentioned product recommendation method.

[0062] According to another aspect of the present invention, there is provided a computer device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0063] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned product recommendation method.

[0064] By means of the above technical solution, the technical solution provided by the embodiment of the present invention has at least the following advantages:

[0065] The present invention provides a product recommendation method and device, storage medium, and computer equipment. First, the subjective feature information of the target user is obtained, and the subjective feature information is extracted based on the consultation dialogue text of the target user; the initial product recommendation result is determined based on the correlation between the subjective feature information and the candidate product combination, and the objective evaluation information of the initial product recommendation result is obtained; the target product recommendation result is generated according to the objective evaluation information, the subjective feature information, and the initial product recommendation result through the trained product recommendation model; and the target product recommendation result is sent to the target user terminal. Compared with the prior art, the embodiment of the present invention extracts subjective feature information based on the dialogue interaction information of the target user, determines the preliminary product recommendation result based on the subjective feature information, and further generates the final product recommendation result according to the objective evaluation information of the preliminary product recommendation result, so as to realize the integration of the user's subjective tendency and the product's objective evaluation in the product recommendation process, so that the recommended product not only meets the user's subjective intention, but also meets the objective market conditions of investment, thereby effectively improving the accuracy of product recommendation.

[0066] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0068] Figure 1 A flow chart of a product recommendation method provided by an embodiment of the present invention is shown;

[0069] Figure 2 A flowchart of another product recommendation method provided by an embodiment of the present invention is shown;

[0070] Figure 3 A block diagram showing the composition of a product recommendation device provided by an embodiment of the present invention is shown;

[0071] Figure 4 A schematic structural diagram of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0072] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0073] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0074] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0075] The existing investment product recommendation is mainly based on matching the collected user information form with the product combination generated by the recommendation technology. However, this method lacks the communication process with the customer and ignores the customer's subjective investment intention. In addition, it also fails to consider the real-time objective market situation, resulting in low accuracy of product recommendation. The embodiment of the present invention provides a product recommendation method, such as Figure 1 As shown, the method is applied to a computer device such as a server as an example for explanation, wherein the server can be an independent server or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN: Content Delivery Network), and basic cloud computing services such as big data and artificial intelligence platforms, such as insurance systems, banking systems, trading systems, etc. The method includes:

[0076] 101. Obtain subjective characteristic information of target users.

[0077] In the embodiment of the present invention, the recommendation of the product is a process that occurs after the product consultation interaction, and the recommendation is realized according to the consultation interaction content of the target user. The specific interaction process can be an interaction in the form of a questionnaire, such as a risk assessment questionnaire, or an interaction in the form of a dialogue, such as a product consultation, which is not specifically limited in the embodiment of the present invention. In some product recommendation scenarios, users usually do not clearly determine the products they need to buy, but only have a preliminary idea of ​​buying. It is necessary to dig out the real needs of the users through communication and interaction to recommend products that meet the needs of the users to the greatest extent. For example, in investment and financial products, the user only expresses the desire to buy stocks, funds, and bonds, but has not selected which product, which needs to be determined through communication. Among them, the communication process can be based on the interaction between service personnel and users, such as financial advisors, fund managers, etc. communicating with users; it can also be based on artificial intelligence Interaction, such as intelligent assistants communicating with users, which is not specifically limited in the embodiment of the present invention.

[0078] It should be noted that since the content generated during the communication and interaction process is not entirely the content that can reflect the user's willingness to buy, when making product recommendations, the subjective characteristic information of the target user is obtained, where the subjective characteristic information is extracted based on the target user's consultation dialogue text. For example, the target user's investment target type, industry of interest, return expectation, risk tolerance, etc. extracted from the consultation dialogue text. By obtaining the subjective characteristic information, the target user's subjective information can be obtained, so that the product recommendation process can fully learn the target user's subjective purchasing tendency, thereby ensuring the matching degree between the recommended product and the target user.

[0079] 102. Determine an initial product recommendation result based on the correlation between the subjective feature information and the candidate product combination, and obtain objective evaluation information of the initial product recommendation result.

[0080] In an embodiment of the present invention, the candidate product combination includes a combination of products currently on sale within the enterprise, and a combination of public offering products on sale in the secondary market. The initial product recommendation result may include multiple product combinations. After obtaining the subjective feature information, a product combination with a high correlation with the subjective feature information is selected from the candidate product combination. Among them, the product combination includes content that the target user is interested in, and may also include content that the target user rejects. For example, if the subjective feature information includes stocks, liquor, funds, and environmental protection, the product combination containing stocks and funds in the liquor and environmental protection industries is determined as the initial product recommendation result. For example, if the subjective feature information includes non-resource stocks, the product combination containing resource stocks in the candidate product combination is added as an excluded product to the initial product recommendation result. Among them, the determination of the relevance can be achieved based on the matching of the product category similarity between the subjective feature information and the candidate product combination, or it can be achieved based on the prediction of the language model, which is not specifically limited in the embodiment of the present invention.

[0081] It should be noted that the initial product recommendation results can only reflect the subjective interest tendencies of the target users, but this recommendation result may not meet the market objectively and may not meet the requirements of the target users. For example, the current market conditions of investment products continue to decline. Therefore, after obtaining the initial product recommendation results, it is also necessary to obtain objective evaluation information of the initial product recommendation results. This objective evaluation information is obtained by combining the actual market status of each product and objective expert knowledge, and can reflect the objective situation of the product. Among them, the objective evaluation information may include evaluation scores for each product or product combination, as well as product-related analysis reports, related information, etc. The above content can be the content obtained by analyzing the product analysis model obtained by training based on the expert knowledge graph, or it can be the analysis result based on the experience of investment consulting service personnel. The embodiment of the present invention does not make specific limitations.

[0082] 103. Generate a target product recommendation result according to the objective evaluation information, the subjective feature information, and the initial product recommendation result through the trained product recommendation model.

[0083] In an embodiment of the present invention, objective evaluation information, subjective feature information, and initial product recommendation results are used as inputs to a trained product recommendation model. The product recommendation model generates a product combination that satisfies both the optimal objective evaluation and the user's interest tendency by learning the correlation between objective evaluation information, subjective feature information, and each product combination. Among them, the target product recommendation result can be one or several combinations of the candidate product combination, or it can be a product combination that is not included in the candidate product combination. That is, the product recommendation model can generate product combinations that did not exist before through learning, so as to break the limitations of existing product combinations, make product configuration plans more flexible, and generate product combinations that can meet the needs of target users to the greatest extent.

[0084] It should be noted that the basic model of the trained product recommendation model is a large-scale language model with a decoder-only structure based on the Transformer framework, such as GPT-2 (Generative Pre-trained Transformer-2), etc., which is not specifically limited in the embodiments of the present invention. The decoder-only model contains only the decoder result, which can not only simplify the model structure, but also combat the low-rank problem of the attention layer, thereby obtaining a better prediction effect.

[0085] 104. Send the target product recommendation result to the target user terminal.

[0086] In an embodiment of the present invention, after obtaining the target product recommendation result, it is sent to the terminal device corresponding to the target user, so that the terminal device displays one or more product combinations included in the target product recommendation result to the target user, and provides a reference for the target user's product now. Among them, the terminal device can be a mobile terminal device, such as a mobile phone, a tablet computer, or a computer, such as a desktop computer, an all-in-one computer, etc., which is not specifically limited in the embodiment of the present invention. The sending form can be a text message, a reply message in the consultation dialogue interface, a message of a chat tool, etc., which is not specifically limited in the embodiment of the present invention.

[0087] In an embodiment of the present invention, for further explanation and limitation, as Figure 2 As shown, the step of determining the initial product recommendation result based on the correlation between the subjective feature information and the candidate product combination includes:

[0088] 201. Using a trained relevance prediction model, predict the candidate product combinations according to the subjective feature information to obtain relevance of different candidate product combinations.

[0089] 202. Sort the candidate product combinations in descending order according to the relevance.

[0090] 203. Generate an initial product recommendation result based on the candidate product combination whose ranking order meets the preset ranking threshold.

[0091] In the embodiment of the present invention, in some scenarios where the number of products is large and the product categories are complex, it is impossible to accurately evaluate the product combination that meets the subjective feature information in the candidate product combination by simply matching the similarity between the subjective feature information and the candidate product combination. Therefore, a large-scale language model with a decoder-only structure of the Transformer framework is used to construct a relevance prediction model. The trained relevance prediction model can predict the relevance of different candidate product combinations relative to the subjective feature information, that is, the rating or score of each candidate product combination. The higher the degree of fit of the candidate product combination with the subjective feature information, the greater the relevance, and the smaller the degree of fit of the candidate product combination, the smaller the relevance. After obtaining the prediction results, the candidate product combinations are sorted according to the relevance, and the corresponding number of candidate product combinations in the sorting column are selected as the initial product recommendation results according to the preset rank threshold. For example, if the preset rank threshold is 3, the candidate product combinations in the top 3 of the ranking are taken as the initial product recommendation results. Among them, the preset rank threshold can be customized according to the specific application scenario, and the embodiment of the present invention does not make specific limitations.

[0092] In one embodiment of the present invention, for further explanation and limitation, before the step of predicting the candidate products according to the consultation dialogue text by using the trained relevance prediction model to obtain the initial recommended products, the method further includes:

[0093] Obtaining a first training sample set, and building an initial relevance prediction model and an initial product recommendation model;

[0094] Using the first training sample set to train the initial relevance prediction model to obtain a trained relevance prediction model;

[0095] Constructing a second training sample set based on the output result of the trained relevance prediction model, wherein the second training sample set includes objective evaluation information samples;

[0096] The initial product recommendation model is trained using the second training sample set to obtain a trained product recommendation model.

[0097] In an embodiment of the present invention, the first training sample set is a sample set of the initial relevance prediction model, which is constructed based on the historical subjective feature information and historical candidate product combinations of global users. The historical subjective feature information can be generated based on the consultation reply model for intelligent reply. Each training sample is marked with the degree of relevance between each candidate product combination and the subjective feature information of the current user. The initial relevance prediction model is trained based on the first training sample, so that the trained relevance prediction model can accurately predict the relevance between each candidate product combination and the consultation feature information, thereby realizing the ranking of the candidate product combinations. After the relevance prediction model is trained, the relevance prediction model is used to predict each first training sample, and the objective evaluation information of the output result of the first training sample is obtained. The subjective feature information in each first training sample, the output result of each first training sample, and the objective evaluation information of the output result are constructed into a second training sample set. The initial product recommendation model is trained according to the second training sample set, so that the trained initial product recommendation model can accurately generate a product combination that meets the subjective feature information and objective evaluation information. The model structure of the relevance prediction model may be the same as or different from that of the product recommendation model, for example, different numbers of decoder layers may be selected, which is not specifically limited in the embodiment of the present invention. The loss function of the initial product recommendation model and the initial relevance prediction model may be a cross entropy function, which is not specifically limited in the embodiment of the present invention.

[0098] It should be noted that by progressive model training and constructing the training sample set of the initial product recommendation model based on the prediction results of the relevance prediction model, the degree of integration between subjective feature information and objective evaluation information can be further improved, thereby improving the accuracy of product combination generation by the product recommendation model.

[0099] In one embodiment of the present invention, for further explanation and limitation, the step of obtaining objective evaluation information of the initial product recommendation result includes:

[0100] Acquire industry information matching the product combination to be recommended from a product industry information database;

[0101] Sending the subjective characteristic information, the initial product recommendation result, and the industry information to an evaluation terminal, so that the evaluation terminal evaluates the product combination to be recommended;

[0102] Receive objective evaluation information of the product combination to be recommended returned by the evaluation terminal.

[0103] In the embodiment of the present invention, the industry information in the product industry information library is industry information related to the product, for example, economic macroeconomic reports, product-related industry development situation reports, quarterly and annual financial reports of product-related companies, etc. Among them, the above-mentioned industry information can be crawled from web pages regularly based on a preset crawler tool, or it can be manually uploaded, and the embodiment of the present invention does not make specific restrictions. In order to facilitate the evaluation of the evaluation terminal, before sending the evaluation request to the evaluation terminal, information that can be used for product evaluation is obtained based on the product combination to be recommended in the initial product recommendation result, and this consultation is sent to the evaluation terminal together with the initial product recommendation result to be evaluated, so as to improve the evaluation efficiency of the evaluation terminal. Among them, the evaluation terminal can be a terminal configured with a product combination evaluation model, which can realize the evaluation and evaluation of the recommended product combination through the learning and analysis of the model, or it can be a terminal device corresponding to the financial service analyst, and the financial service analyst analyzes, scores, adds explanatory notes, etc. to the recommended product combination one by one based on the received information, and the embodiment of the present invention does not make specific restrictions.

[0104] It should be noted that the objective evaluation information replied by the evaluation terminal includes at least one of the recommendation parameters and the evaluation reference factors. Among them, the recommendation parameters may be the recommendation score, recommendation coefficient, etc. of the product combination to be recommended, and the evaluation reference factors may be explanatory notes, analytical comments, relevant reference materials used to support the evaluation, etc., which are not specifically limited in the embodiments of the present invention. Obtaining the objective evaluation information of the product combination to be recommended can achieve an objective analysis of the target user's subjective preference for products and play a professional product guidance role. For example, the target user tends to buy liquor stocks, but after objective analysis, the liquor target is not in a good investment range, but the fund of funds (FOF, Fund of Funds) can be considered, and an analysis report can be attached to achieve an effective fusion of subjective and objective information and ensure the accuracy of product combination recommendations.

[0105] In one embodiment of the present invention, for further explanation and limitation, before the step of obtaining the subjective characteristic information of the target user, the method further includes:

[0106] Receive the inquiry question text of the target user;

[0107] Generate a reply text to the consulting question text according to the consulting question text by using the trained consulting reply model;

[0108] Sending the reply text to the terminal of the target user to obtain the next consultation question text of the target user;

[0109] In response to an end instruction indicating the end of the consultation with the target user, subjective feature information is generated according to the consultation question text through the trained consultation response model.

[0110] In an embodiment of the present invention, when product recommendations are made based on the interactive content of a consulting dialogue, the interactive dialogue process can respond to consulting questions raised by target users based on a consulting response model. The trained consulting response model is trained based on the communication content between investment service personnel, such as fund managers, and clients as training samples. The trained consulting response model can output corresponding response information for questions raised by users, and can generate subjective feature information for summarizing the content of the dialogue and extracting key information at the end of the conversation. The basic model of this consulting response model can be the same as the model structure of the product recommendation model or the relevance prediction model, or it can be different, for example, by selecting a different number of Decoder layers, which is not specifically limited in the embodiment of the present invention.

[0111] It should be noted that product recommendations are aimed at a large number of users. If they are completely based on manual work, they can only meet a small number of user needs. For example, for existing investment and financial products, most of them can only enjoy the consulting services and product recommendations of fund managers as one-on-one advanced users, while most financial investment users cannot enjoy consulting services and can only obtain product recommendation services by filling out risk questionnaires. By replacing manual services with consultation response models, the efficiency of consulting services and the audience of consulting services can be greatly improved, while at the same time, the labor cost of product recommendations can be reduced. Since subsequent product recommendations also need to integrate objective evaluation information, the consultation process is only an exploration of subjective information. Therefore, answering user inquiries based on the consultation response model and extracting consultation feature information can still ensure the accuracy of product recommendations.

[0112] In one embodiment of the present invention, for further explanation and limitation, before the step of obtaining the subjective characteristic information of the target user, the method further includes:

[0113] Obtaining the consultation dialogue text of the target user;

[0114] Feature information is extracted from the user-side dialogue text in the consultation dialogue text using a trained feature extraction model to obtain subjective feature information.

[0115] In the embodiment of the present invention, the consultation dialogue text is generated based on the speech recognition model, that is, the current consultation scenario is a scenario of consultation based on manual telephone service. When the target user consults the product based on the consultation call, the recording of the telephone interaction is recorded, and the telephone recording is converted into the consultation dialogue text based on the speech recognition model, and then the feature extraction model that has been trained is used to extract feature information from the consultation dialogue text to obtain subjective feature information. For example, during the telephone communication process, the customer mentioned "I feel that the pharmaceutical industry is better" and "What will the return be if I buy ETF products?", etc., then the extracted subjective feature information includes "pharmaceutical industry" and "ETF products". Among them, the basic model of the feature extraction model is a pre-trained semantic model, such as GPT, GPT-2 and other models, which are not specifically limited in the embodiment of the present invention. Among them, the training sample of the feature extraction model is the user-side dialogue text marked with subjective key information. Based on the training of this training sample, the self-attention in the multi-layer Transformer can fully learn the content related to the user's subjective tendency in the consultation dialogue text, thereby realizing the accurate extraction of subjective feature information.

[0116] In one embodiment of the present invention, for further explanation and limitation, the target product recommendation result is any candidate product combination in the initial product recommendation result or a recombined product combination different from the candidate product combination, and the recombined product combination is the recombined result of at least two single products in the initial product recommendation combination.

[0117] In an embodiment of the present invention, the target product recommendation result can be any candidate product combination, or it can be a combination result obtained by recombining two or more single products in the initial product recommendation combination, which is different from any candidate product combination. For example, the initial product recommendation combination includes a combination of products on sale and a combination of publicly offered products, wherein the combination of products on sale includes internal product A and internal product B, and the combination of publicly offered products includes publicly offered products C and publicly offered products D. After the product recommendation model learns objective evaluation information and subjective feature information, the generated target product recommendation result is a product combination consisting of internal product A and publicly offered product C. The combination form of this product combination is different from any existing candidate product combination. Based on objective evaluation information and subjective feature information, the current product combination is shuffled and rearranged in units of single products to avoid product recommendations being restricted by existing product combinations, so that the target product recommendation results are highly matched with the needs of target users, thereby greatly improving the accuracy and flexibility of product recommendations.

[0118] In an embodiment of the present invention, in order to ensure that the candidate product combination is comprehensive and accurate, the obtained target product recommendation results are matched with the candidate product combination. If the matching result is that the target product recommendation results include a product combination that is not in the candidate product combination, that is, the product combination is rearranged, then this product combination is updated to the candidate product combination so that the subsequent user's product recommendation process can be based on the latest candidate product combination for recommendation.

[0119] The present invention provides a product recommendation method, firstly obtaining subjective characteristic information of a target user, wherein the subjective characteristic information is extracted based on the consultation dialogue text of the target user; determining an initial product recommendation result based on the correlation between the subjective characteristic information and a candidate product combination, and obtaining objective evaluation information of the initial product recommendation result; generating a target product recommendation result based on the objective evaluation information, the subjective characteristic information, and the initial product recommendation result through a trained product recommendation model; and sending the target product recommendation result to a target user terminal. Compared with the prior art, the embodiment of the present invention extracts subjective characteristic information based on the dialogue interaction information of the target user, determines a preliminary product recommendation result based on the subjective characteristic information, and further generates a final product recommendation result based on the objective evaluation information of the preliminary product recommendation result, thereby realizing the integration of the user's subjective tendency and the product's objective evaluation in the product recommendation process, so that the recommended product not only satisfies the user's subjective intention, but also satisfies the objective market conditions of the investment, thereby effectively improving the accuracy of product recommendations.

[0120] Furthermore, as a response to the above Figure 1 The implementation of the method shown in the figure, the embodiment of the present invention provides a product recommendation device, such as Figure 3 As shown, the device comprises:

[0121] According to another aspect of the present invention, a product recommendation device is provided, comprising:

[0122] An acquisition module 31 is used to acquire subjective characteristic information of a target user, wherein the subjective characteristic information is extracted based on a consultation dialogue text of the target user;

[0123] A determination module 32, configured to determine an initial product recommendation result based on the correlation between the subjective feature information and the candidate product combination, and obtain objective evaluation information of the initial product recommendation result;

[0124] A prediction module 33 is used to generate a target product recommendation result according to the objective evaluation information, the subjective feature information, and the initial product recommendation result through a trained product recommendation model;

[0125] The sending module 34 is used to send the target product recommendation result to the target user terminal.

[0126] Further, the determination module 32 includes:

[0127] A prediction unit, configured to perform prediction processing on the candidate product combinations according to the subjective feature information by using a trained relevance prediction model, so as to obtain relevances of different candidate product combinations;

[0128] A sorting unit, used to sort the candidate product combinations in descending order according to the relevance;

[0129] The generating unit is used to generate an initial product recommendation result based on a combination of candidate products whose rankings meet a preset ranking threshold.

[0130] Furthermore, the device also includes:

[0131] A first construction module is used to obtain a first training sample set and construct an initial relevance prediction model and an initial product recommendation model;

[0132] A first training module, used to train the initial relevance prediction model using the first training sample set to obtain a trained relevance prediction model;

[0133] A second construction module is used to construct a second training sample set based on the output result of the trained relevance prediction model, wherein the second training sample set includes objective evaluation information samples;

[0134] The second training module is used to train the initial product recommendation model using the second training sample set to obtain a trained product recommendation model.

[0135] Furthermore, the determining module 32 further includes:

[0136] An acquisition unit, used for acquiring industry information matching the product combination to be recommended from a product industry information database;

[0137] A sending unit, configured to send the subjective characteristic information, the initial product recommendation result, and the industry information to an evaluation terminal, so that the evaluation terminal evaluates the product combination to be recommended;

[0138] The first receiving unit is configured to receive objective evaluation information of the to-be-recommended product combination returned by the evaluation terminal, wherein the objective evaluation information includes at least one of a recommendation parameter and an evaluation reference factor.

[0139] Furthermore, the device also includes:

[0140] A receiving module, used for receiving the consulting question text of the target user;

[0141] A first generating module is used to generate a reply text of the consulting question text according to the consulting question text by using the trained consulting reply model;

[0142] The sending module 34 is further used to send the reply text to the terminal of the target user to obtain the next consultation question text of the target user;

[0143] The second generating module is used to generate subjective feature information according to the consultation question text through the trained consultation response model in response to an end instruction indicating the end of the consultation of the target user.

[0144] Furthermore, the device also includes:

[0145] The acquisition module 31 is further used to acquire the consultation dialogue text of the target user, wherein the consultation dialogue text is generated based on the speech recognition model;

[0146] The extraction module is used to extract feature information from the user-side dialogue text in the consultation dialogue text through a trained feature extraction model to obtain subjective feature information.

[0147] Furthermore, the device also includes:

[0148] An updating module is used for updating the target recommended product combination into the candidate product combination if the candidate product combination does not include the target recommended product combination.

[0149] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to execute operations corresponding to the above-mentioned product recommendation method.

[0150] According to another aspect of the present invention, there is provided a computer device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0151] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned product recommendation method.

[0152] The present invention provides a product recommendation device, which first obtains the subjective characteristic information of a target user, wherein the subjective characteristic information is extracted based on the consultation dialogue text of the target user; determines the initial product recommendation result based on the correlation between the subjective characteristic information and the candidate product combination, and obtains the objective evaluation information of the initial product recommendation result; generates the target product recommendation result according to the objective evaluation information, the subjective characteristic information, and the initial product recommendation result through the trained product recommendation model; and sends the target product recommendation result to the target user terminal. Compared with the prior art, the embodiment of the present invention extracts the subjective characteristic information based on the dialogue interaction information of the target user, determines the preliminary product recommendation result based on the subjective characteristic information, and further generates the final product recommendation result according to the objective evaluation information of the preliminary product recommendation result, thereby realizing the fusion of the user's subjective tendency and the product's objective evaluation in the product recommendation process, so that the recommended product not only satisfies the user's subjective intention, but also satisfies the objective market conditions of the investment, thereby effectively improving the accuracy of product recommendation.

[0153] According to an embodiment of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the computer executable instruction can execute the product recommendation method in any of the above method embodiments.

[0154] Figure 4 A schematic diagram of the structure of a computer device provided according to an embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computer device.

[0155] like Figure 4 As shown, the computer device may include: a processor (processor) 402 , a communication interface (Communications Interface) 404 , a memory (memory) 406 , and a communication bus 408 .

[0156] The processor 402 , the communication interface 404 , and the memory 406 communicate with each other via a communication bus 408 .

[0157] The communication interface 404 is used for network communication with other devices such as a client or other servers.

[0158] The processor 402 is used to execute the program 410, and specifically can execute the relevant steps in the above-mentioned product recommendation method embodiment.

[0159] Specifically, the program 410 may include program codes, which include computer operation instructions.

[0160] The processor 402 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computer device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0161] The memory 406 is used to store the program 410. The memory 406 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0162] The program 410 may be specifically configured to enable the processor 402 to perform the following operations:

[0163] Obtaining subjective characteristic information of a target user, wherein the subjective characteristic information is extracted based on a consultation dialogue text of the target user;

[0164] Determine an initial product recommendation result based on the relevance between the subjective feature information and the candidate product combination, and obtain objective evaluation information of the initial product recommendation result;

[0165] Generate a target product recommendation result according to the objective evaluation information, the subjective feature information, and the initial product recommendation result through the trained product recommendation model;

[0166] The target product recommendation result is sent to the target user terminal.

[0167] Obviously, it should be understood by those skilled in the art that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that herein, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0168] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A product recommendation method, characterized in that: include: Obtaining subjective characteristic information of a target user, wherein the subjective characteristic information is extracted based on a consultation dialogue text of the target user; Determine an initial product recommendation result based on the relevance between the subjective feature information and the candidate product combination, and obtain objective evaluation information of the initial product recommendation result; Generate a target product recommendation result according to the objective evaluation information, the subjective feature information, and the initial product recommendation result through the trained product recommendation model; The target product recommendation result is sent to the target user terminal.

2. The method according to claim 1, characterized in that The determining of the initial product recommendation result based on the correlation between the subjective feature information and the candidate product combination includes: Using the trained relevance prediction model, predicting the candidate product combinations according to the subjective feature information, and obtaining the relevance of different candidate product combinations; Sorting the candidate product combinations in descending order according to the relevance; Generate initial product recommendation results based on candidate product combinations whose rankings meet a preset ranking threshold.

3. The method according to claim 2, characterized in that Before the candidate products are predicted and processed according to the consultation dialogue text by using the trained relevance prediction model to obtain the initial recommended product results, the method further includes: Obtaining a first training sample set, and building an initial relevance prediction model and an initial product recommendation model; Using the first training sample set to train the initial relevance prediction model to obtain a trained relevance prediction model; Constructing a second training sample set based on the output result of the trained relevance prediction model, wherein the second training sample set includes objective evaluation information samples; The initial product recommendation model is trained using the second training sample set to obtain a trained product recommendation model.

4. The method according to claim 1, characterized in that: The initial product recommendation result includes at least one product combination to be recommended, and the obtaining of objective evaluation information of the initial product recommendation result includes: Acquire industry information matching the product combination to be recommended from a product industry information database; Sending the subjective characteristic information, the initial product recommendation result, and the industry information to an evaluation terminal, so that the evaluation terminal evaluates the product combination to be recommended; Receive objective evaluation information of the to-be-recommended product combination returned by the evaluation terminal, wherein the objective evaluation information includes at least one of a recommendation parameter and an evaluation reference factor.

5. The method according to claim 1, characterized in that Before obtaining the subjective characteristic information of the target user, the method further includes: Receive the inquiry question text of the target user; Generate a reply text to the consulting question text according to the consulting question text by using the trained consulting reply model; Sending the reply text to the terminal of the target user to obtain the next consultation question text of the target user; In response to an end instruction indicating the end of the consultation with the target user, subjective feature information is generated according to the consultation question text through the trained consultation response model.

6. The method according to claim 1, characterized in that Before obtaining the subjective characteristic information of the target user, the method further includes: Acquire a consultation dialogue text of the target user, where the consultation dialogue text is generated based on a speech recognition model; Feature information is extracted from the user-side dialogue text in the consultation dialogue text using a trained feature extraction model to obtain subjective feature information.

7. The method according to any one of claims 1 to 6, characterized in that The target product recommendation result is any candidate product combination in the initial product recommendation result or a recombined product combination different from the candidate product combination, and the recombined product combination is a recombined result of at least two single products in the initial product recommendation combination.

8. A product recommendation device, characterized in that: include: An acquisition module, used to acquire subjective characteristic information of a target user, wherein the subjective characteristic information is extracted based on a consultation dialogue text of the target user; A determination module, configured to determine an initial product recommendation result based on the relevance between the subjective feature information and the candidate product combination, and obtain objective evaluation information of the initial product recommendation result; A prediction module, configured to generate a target product recommendation result according to the objective evaluation information, the subjective feature information, and the initial product recommendation result through a trained product recommendation model; The sending module is used to send the target product recommendation result to the target user terminal.

9. A storage medium, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to perform operations corresponding to the product recommendation method as described in any one of claims 1-7.

10. A computer device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the product recommendation method according to any one of claims 1-7.

Images