Product recommendation method and device, equipment, medium and product

Abstract

The embodiment of the invention provides a product recommendation method and device, equipment, a medium and a product, and relates to the field of artificial intelligence or the field of financial science and technology. Firstly, a product recommendation request of a user is received, and user identification information is obtained. And then, based on the user identification information, determining a corresponding intention identification result. Thirdly, generating a plurality of recommended products and recommendation reasons corresponding to the recommended products according to the intention recognition result through a collaborative recommendation agent group; and finally, outputting the recommendation results to the user. According to the method, through flexible collaboration and intention deep recognition of the collaborative recommendation agent group, the limitation of a traditional fixed rule is broken through, and accurate recommendation meeting the personalized requirements of the user is realized; meanwhile, the interpretability and the credibility of the recommendation result are improved through the presentation mode with the recommendation reason, the user interaction experience is enhanced, and the product recommendation adaptation degree and the user acceptability are effectively improved.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence or fintech, and more particularly to a product recommendation method, apparatus, device, medium, and product. Background Technology

[0002] Intelligent customer service systems are important tools in banking operations for improving customer service efficiency. Through intelligent customer service, customers can easily interact with the system via mobile banking apps or websites to receive personalized service recommendations.

[0003] Existing product recommendation methods primarily rely on rule engines or simple FAQ matching mechanisms. Essentially, they use pre-defined fixed business rules and keyword matching templates to perform surface-level parsing of the user's input request text, and then retrieve products from the product library that meet the basic criteria for recommendation.

[0004] However, this recommendation method, which relies on rule engines or simple FAQ matching, cannot fully consider users' personalized needs, preferences, and investment backgrounds because it is mainly based on fixed rules or templates. As a result, the recommended financial products often fail to meet users' diverse needs. Summary of the Invention

[0005] This application provides product recommendation methods, apparatus, devices, media, and products to address the problem that existing product recommendation methods rely on fixed rules or templates and cannot meet the diverse needs of users.

[0006] In a first aspect, embodiments of this application provide a product recommendation method, including:

[0007] Receive a product recommendation request, the product recommendation request including: user identification information;

[0008] Determine the intent recognition result corresponding to the user identification information;

[0009] By using a collaborative recommendation intelligent agent group, a product recommendation result corresponding to the intent recognition result is determined. The product recommendation result includes: multiple recommended products and the recommendation reason for each recommended product.

[0010] Output the product recommendation results.

[0011] In one possible implementation, the collaborative recommendation agent group includes: a first agent, a second agent, and a third agent. The step of determining the product recommendation result corresponding to the intent recognition result through the collaborative recommendation agent group includes:

[0012] The first intelligent agent performs anomaly assessment on the user profile associated with the user identification information and generates anomaly level.

[0013] The second intelligent agent determines the multiple recommended products based on the anomaly level and the intent recognition result.

[0014] The third intelligent agent generates recommendation reasons for each of the multiple recommended products, and determines the product recommendation result based on each recommended product and its corresponding recommendation reasons.

[0015] In one possible implementation, the collaborative recommendation agent group further includes: a fourth agent, and the method further includes:

[0016] If the user identification information indicates that the user is a new user, then the fourth intelligent agent will conduct a question-and-answer interaction with the new user to obtain the question-and-answer results, and based on the question-and-answer results, determine the user profile of the new user;

[0017] If the user identification information indicates that the user is a previous user, then based on the user identification information, historical transaction data is obtained, and the historical transaction data is analyzed and processed to obtain a user profile of the previous user.

[0018] In one possible implementation, determining the plurality of recommended products by the second intelligent agent based on the anomaly level and the intent recognition result includes:

[0019] The second intelligent agent obtains a product knowledge graph, which includes multiple candidate products and product attributes corresponding to each candidate product.

[0020] Based on the product attributes corresponding to each of the candidate products, the product level corresponding to each of the candidate products is determined;

[0021] Based on the intent recognition result, the anomaly level, and the product level corresponding to each candidate product, the multiple recommended products are determined from the multiple candidate products.

[0022] In one possible implementation, the step of generating recommendation reasons for each of the multiple recommended products through the third intelligent agent includes:

[0023] The third intelligent agent inputs the product attributes corresponding to each recommended product into the language generation model to obtain the recommendation reasons for each recommended product.

[0024] In one possible implementation, the step of engaging in question-and-answer interaction with the new user through the fourth intelligent agent to obtain the question-and-answer results includes:

[0025] During the question-and-answer interaction, the fourth intelligent agent obtains a preset filter word list;

[0026] Receive a first user question, and filter the first user question based on the filter vocabulary to obtain a second user question;

[0027] Based on the second user's question, a question-and-answer interaction is conducted with the user until the question-and-answer result is obtained.

[0028] In one possible implementation, the method further includes:

[0029] Acquire multimodal data;

[0030] The multimodal data is input into the behavior feature analysis model to obtain the behavior feature results output by the behavior feature analysis model;

[0031] Based on the behavioral characteristics, the product recommendation results are adjusted to obtain the adjusted product recommendation results, and the adjusted product recommendation results are output.

[0032] Secondly, embodiments of this application provide a product recommendation device, comprising:

[0033] The receiving module is used to receive product recommendation requests, wherein the product recommendation requests include: user identification information;

[0034] The determination module is used to determine the intent recognition result corresponding to the user identification information;

[0035] The determining module is further configured to determine the product recommendation result corresponding to the intent recognition result through a collaborative recommendation intelligent agent group, wherein the product recommendation result includes: multiple recommended products and the recommendation reason corresponding to each recommended product;

[0036] The output module is used to output the product recommendation results.

[0037] In one possible implementation, the apparatus further includes: a generation module;

[0038] The generation module is used to perform anomaly assessment on the user profile associated with the user identification information through the first intelligent agent and generate anomaly level;

[0039] The determining module is specifically used to determine the multiple recommended products based on the anomaly level and the intent recognition result through the second intelligent agent;

[0040] The determining module is specifically used by the third intelligent agent to generate recommendation reasons for each of the multiple recommended products, and to determine the product recommendation result based on each of the recommended products and the recommendation reasons for each of the recommended products.

[0041] In one possible implementation, if the user identification information indicates that the user is a new user, the determining module is further configured to engage in question-and-answer interaction with the new user through the fourth intelligent agent to obtain the question-and-answer results, and determine the user profile of the new user based on the question-and-answer results;

[0042] The device further includes: an acquisition module;

[0043] If the user identification information indicates that the user is a previous user, the acquisition module is used to acquire historical transaction data based on the user identification information.

[0044] The device further includes: a processing module;

[0045] The processing module is used to analyze and process the historical transaction data to obtain the user profile of the old user.

[0046] In one possible implementation, the acquisition module is further configured to acquire a product knowledge graph through the second intelligent agent, the product knowledge graph including multiple candidate products and product attributes corresponding to each candidate product;

[0047] The determining module is further configured to determine the product level corresponding to each candidate product based on the product attributes corresponding to each candidate product;

[0048] The determining module is specifically used to determine the multiple recommended products from the multiple candidate products based on the intent recognition result, the anomaly level, and the product level corresponding to each candidate product.

[0049] In one possible implementation, the device further includes: an input module;

[0050] The input module is used to input the product attributes corresponding to each of the recommended products into the language generation model through the third intelligent agent to obtain the recommendation reasons corresponding to each of the recommended products.

[0051] In one possible implementation, the acquisition module is further configured to acquire a preset filter word list through the fourth intelligent agent during the question-and-answer interaction process.

[0052] The receiving module is also used to receive a first user question;

[0053] The processing module is further configured to filter the first user question based on the filter vocabulary to obtain the second user question;

[0054] The determining module is specifically used to engage in question-and-answer interaction with the user based on the second user's question until the question-and-answer result is obtained.

[0055] In one possible implementation, the acquisition module is further configured to acquire multimodal data;

[0056] The input module is also used to input the multimodal data into the behavior feature analysis model to obtain the behavior feature results output by the behavior feature analysis model;

[0057] The processing module is further configured to adjust the product recommendation result based on the behavioral feature result, obtain the adjusted product recommendation result, and output the adjusted product recommendation result.

[0058] Thirdly, embodiments of this application provide a product recommendation device, including: a memory and a processor;

[0059] The memory stores computer-executed instructions;

[0060] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0061] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0062] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0063] The product recommendation method, apparatus, device, medium, and product provided in this application first receive a product recommendation request containing user identification information, then determine the user intent recognition result based on the identification information, and then call a collaborative recommendation intelligent agent group to generate product recommendation results matching the intent, finally outputting the product recommendation results. This method, through the collaborative operation of the intelligent agent group, effectively solves the limitations of traditional recommendation systems that rely on fixed rules or preset templates. It can dynamically analyze diverse user needs and generate explanatory recommendation schemes, solving the recommendation bias problem caused by rigid rules in the prior art, and enhancing the credibility of user decisions through the explicit presentation of recommendation reasons, thereby improving the reliability of the recommendation system and user satisfaction. Attached Figure Description

[0064] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0065] Figure 1 Flowchart of the product recommendation method provided in this application Figure 1 ;

[0066] Figure 2 Flowchart of the product recommendation method provided in this application Figure 2 ;

[0067] Figure 3 A schematic diagram of the product recommendation device provided in this application;

[0068] Figure 4 A schematic diagram of the structure of the recommended equipment for the product provided in this application.

[0069] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0070] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0071] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation access points for users to choose to authorize or refuse.

[0072] Furthermore, the technical solution involved in this application, which involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.) and the use of artificial intelligence technology for automated decision-making, and makes decisions that have a significant impact on personal rights based on the results of automated decision-making, provides users with corresponding operation entry points for users to choose to agree to or reject the results of automated decision-making; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0073] It should be noted that the product recommendation methods, apparatus, equipment, media and products provided in this application can be used in the fields of artificial intelligence or fintech, or in any field other than artificial intelligence or fintech. The application fields of the product recommendation methods, apparatus, equipment, media and products in this application are not limited.

[0074] Intelligent customer service systems are important tools in banking operations for improving customer service efficiency. Through intelligent customer service, customers can easily interact with the system via mobile banking apps or websites to obtain personalized recommendations. For example, a customer might access the intelligent customer service interface through a mobile banking app or website and make requests such as "I want a stable-return investment product" or "Help me select suitable money market funds."

[0075] Existing methods for recommending financial products primarily rely on rule engines or simple FAQ matching mechanisms. Essentially, they use pre-defined fixed business rules and keyword matching templates to perform a superficial parsing of the user's input request text, and then retrieve products from a product database that meet the basic criteria for recommendation.

[0076] However, this recommendation method, which relies on rule engines or simple FAQ matching, cannot fully consider users' personalized needs, preferences, and investment backgrounds because it is mainly based on fixed rules or templates. As a result, the recommended financial products often fail to meet users' diverse needs.

[0077] The product recommendation method provided in this application first receives a product recommendation request containing user identification information, then determines the user intent based on the identification information, and subsequently calls a collaborative recommendation agent group to generate product recommendation results that match the intent, finally outputting the product recommendation results. This method, through the collaborative operation of the agent group, effectively solves the limitations of traditional recommendation systems that rely on fixed rules or preset templates. It can dynamically analyze diverse user needs and generate explanatory recommendation schemes, thus solving the recommendation bias problem caused by rigid rules in existing technologies. Furthermore, by explicitly presenting the reasons for recommendations, it enhances the credibility of user decisions, achieving a technological leap from data matching to understanding needs, and significantly improving the reliability and user satisfaction of the recommendation system.

[0078] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0079] Figure 1 Flowchart of the product recommendation method provided in this application Figure 1 ,like Figure 1 As shown, the method includes:

[0080] S101. Receive a product recommendation request. The product recommendation request includes: user identification information.

[0081] The user identification information is used to uniquely identify each user. This user identification information can be, for example, a user's account or a user's mobile phone number. This application does not impose any special restrictions on this.

[0082] The purpose of this step is to receive product recommendation requests from users and obtain user identification information.

[0083] Understandably, this is because the system needs to receive product recommendation requests to perceive users' proactive needs. Without receiving such requests, the system cannot determine whether a user requires the recommendation service, and therefore subsequent recommendation operations cannot be initiated. Thus, it is essential to first receive product recommendation requests to ascertain user needs and then provide the necessary recommendation services.

[0084] This step involves receiving product recommendation requests, for example, through designated interactive entry points such as bank mobile applications, web-based wealth management sections, and mini-programs, where users can click on the relevant function button to recommend wealth management products to send a product recommendation request to the system. This application does not impose any special restrictions on this.

[0085] S102. Determine the intent recognition result corresponding to the user identification information.

[0086] The intent recognition result refers to the user's current needs inferred from their identification information. For example, if the product recommendation request is "Please recommend low-risk financial products," then based on the above information, the intent recognition result can be determined to be a low-risk product recommendation.

[0087] The purpose of this step is to determine the intent recognition result associated with the user based on the user identification information.

[0088] It's understandable that different users have different needs and behaviors. Only by identifying the user's intent can product recommendations be made based on their personalized needs. Without clear intent identification results, irrelevant or uninteresting products may be offered, leading to poor recommendation performance and a degraded user experience.

[0089] S103. Through a collaborative recommendation agent group, determine the product recommendation results corresponding to the intent recognition results. The product recommendation results include: multiple recommended products and the recommendation reasons for each recommended product.

[0090] The collaborative recommendation agent group includes, but is not limited to, a first agent, a second agent, a third agent, and a fourth agent. Each agent has a clearly defined role and works in concert: the first agent assesses and determines the user's anomaly level, providing a basis for risk and suitability verification for product recommendations; the second agent, based on the user's intent recognition results and the aforementioned anomaly level, filters and determines eligible recommended products; the third agent generates corresponding recommendation reasons for the recommended products output by the second agent, improving the interpretability of the recommendation results; and the fourth agent engages in dialogue with the user to further obtain or supplement user needs information, assisting in optimizing the recommendation process.

[0091] The generation of collaborative recommendation agent groups adopts a training strategy of "first training a single agent specifically, then optimizing multiple agents collaboratively." This strategy includes: in the single agent design phase, for the first agent, a risk assessment model is trained based on historical data (including user profiles of different historical users in terms of occupation, age, assets, customer star rating, financial risk preference, held financial products, and historical browsing records); for the second agent, a product recommendation model is trained based on the anomaly level output by the first agent and the user intent recognition results; for the third agent, a preference suggestion generation model is trained based on the recommended product-related data selected by the second agent; and for the fourth agent, a question-and-answer matching and generation model is trained based on question-and-answer datasets from the financial field and historical dialogues.

[0092] In the multi-agent collaborative training phase, firstly, a simulated scenario encompassing the entire user service process is constructed. This process includes: user initiating a financial management request → risk assessment → product recommendation → reason generation → user questioning → Q&A interaction. Subsequently, the collaborative process of each agent revolves around this entire process: when a simulated user initiates a financial management request, the first agent is triggered to complete an anomaly assessment of the user profile. The assessment result is then synchronized to the second agent, which combines the user intent recognition result to filter recommended products. Next, the second agent transfers the recommended product data to the third agent, generating recommendation reasons tailored to user preferences and forming a product recommendation. If a simulated user raises a question regarding the product recommendation results, the fourth agent receives the question in real time and can access intermediate data from the first three agents (such as product details and user preference tags) to respond until a clear question-and-answer result is output. Subsequently, each specially trained individual agent is integrated into the collaborative training framework as a functional module. The interaction logic between the agents is optimized through reinforcement learning or multi-agent game algorithms (such as the matching degree between risk level and product recommendation, and the fit between recommendation reasons and user preferences). The model parameters of each agent are continuously iterated based on real user feedback data to form a group of recommendation agents with collaborative capabilities.

[0093] In the training process of the first agent, the training algorithm employs either logistic regression or random forest. Logistic regression uses a linear model to fit the data distribution, thereby classifying and judging the anomaly level. Random forest improves the stability and accuracy of the assessment by integrating the prediction results of multiple decision trees. The base model used for training is a risk assessment-specific model. The training input consists of historical data and multi-dimensional user profile data (including occupation, age, assets, etc.). The training output is the anomaly level corresponding to the user profile. During training, the input data is preprocessed (e.g., feature encoding, missing value imputation) before being fed into the base model. The model parameters are optimized using methods such as gradient descent (an optimization method that continuously adjusts model parameters to reduce prediction error) and cross-validation (multiple training and validations by dividing the dataset to avoid overfitting). This iterative process continues until the model meets the assessment criteria for anomaly levels, ultimately achieving an accurate assessment of the anomaly level of the user profile.

[0094] During the training of the second agent, the training algorithm employs collaborative filtering, content-based recommendation, or a hybrid recommendation algorithm combining both. Collaborative filtering recommends products favored by similar users by analyzing user groups with similar behaviors or preferences. Content-based recommendation filters products that meet user needs based on the matching degree between product characteristics (such as risk level, return type, and investment period) and user requirements. The hybrid recommendation algorithm combines the advantages of the first two algorithms, considering both user group preferences and direct matching between products and needs, thus improving the accuracy and comprehensiveness of recommendations. The basic model used for training is a product recommendation model. The training input consists of the anomaly level and user intent recognition results output by the first agent; the training output is a list of recommended products that meet the criteria. During training, the input data is combined with massive amounts of historical recommendation data and fed into the model. The model parameters are continuously optimized through iterative iteration to ensure that the recommended products output by the model match the user's risk tolerance and core needs.

[0095] During the training of the third agent, the training algorithms directly employ generative pre-trained language model algorithms derived from the Transformer architecture and encoder-decoder architecture generative algorithms. The base model used for training is a preference recommendation reason generation model. The training input consists of recommended product information (such as product type, risk level, and return) determined by the second agent and preference tags from the user profile. The training output is personalized recommendation reasons that match user preferences. Specifically, the generative pre-trained language model algorithm based on the Transformer architecture learns the rules of language expression and semantic relationships, while capturing semantic dependencies in long texts to improve the coherence of the recommendation reason expression; the encoder-decoder architecture generative algorithm ensures that the recommendation reasons conform to product characteristics and user preferences by matching the semantic correspondence between the input product information, user preference tags, and output text. During training, high-quality recommendation reasons annotated manually are used as a reference standard. An "unsupervised pre-training + supervised fine-tuning" approach is adopted, inputting the input data into the base model for training, continuously optimizing the model's language generation capabilities, and ensuring that the output recommendation reasons conform to user preferences.

[0096] In the training process of the fourth agent, the training algorithm employs question-answering matching algorithms (such as semantic matching algorithms based on pre-trained language models like BERT and RoBERTa) and generative question-answering algorithms. The semantic matching algorithm based on pre-trained language models like BERT and RoBERTa primarily learns the semantic features of text through the model, calculates the semantic similarity between the user's question and the standard questions in the financial domain question-answering dataset, and quickly locates the most matching standard answer. The generative question-answering algorithm learns the generation logic of natural language and the context of dialogue, enabling it to autonomously generate natural language answers that conform to semantic logic and accurately respond to user questions that do not match the standard answer in the question-answering dataset. The basic model used for training is a question-answering interaction model (including a matching module and a generation module). The training input is divided into two parts: the matching module input is the financial domain question-answering dataset, and the generation module input is the historical dialogue dataset. The training output also corresponds to two parts: the matching module output is the matching result between the user's question and the standard question, and the generation module output is the natural language answer when no standard answer is matched. The training aims to improve question-answering accuracy and user satisfaction. Two modules are trained separately, while the collaborative logic between the modules is optimized in conjunction with the training. The model knowledge base is continuously enriched to ensure that it can respond to various user questions.

[0097] The purpose of this step is to translate user intent into actionable product recommendation solutions.

[0098] Understandably, firstly, since user intent recognition results include multi-dimensional needs, a single module cannot fully cover complex tasks such as screening, evaluation, and reason generation. However, multi-agent collaboration can improve processing efficiency and accuracy through division of labor and cooperation, breaking through the limitations of traditional single-rule recommendations.

[0099] Secondly, the recommendation results provide multiple products and corresponding reasons, which can not only meet the diverse preferences of users in different scenarios, but also solve the shortcomings of traditional recommendations that only provide results without evidence by clearly explaining product suitability, thereby enhancing user trust and increasing recommendation acceptance rate.

[0100] S104. Output product recommendation results.

[0101] The purpose of this step is to present the product recommendation results generated by the collaborative recommendation agent group to the user clearly and accurately.

[0102] Understandably, by providing product recommendation results, we can not only show users products that meet their requirements, but also help them quickly understand the recommendation logic through the reasons for the recommendations. This will improve the transparency and trustworthiness of the recommendations, enhance user engagement and satisfaction, and improve the overall user experience.

[0103] The product recommendation method provided in this application first receives a product recommendation request initiated by a user, the request including user identification information; then, based on the user identification information, accurately determines the corresponding user intent recognition result, clarifying the user's core financial needs and potential preferences; next, through the collaborative work of multiple modules of the collaborative recommendation intelligent agent group, combined with the intent recognition result, suitable products are selected and recommendation reasons for each product are generated, ultimately forming a product recommendation result containing multiple recommended products and corresponding recommendation reasons; finally, the product recommendation result is fed back to the user.

[0104] This method overcomes the limitations of traditional fixed rules by leveraging the flexible collaboration and deep intent recognition of collaborative recommendation agents, achieving accurate recommendations tailored to users' personalized needs. Simultaneously, the presentation of accompanying recommendation reasons enhances the interpretability and credibility of the recommendation results, strengthens the user interaction experience, and effectively improves the adaptability and user acceptance of product recommendations.

[0105] Figure 2 Flowchart of the product recommendation method provided in this application Figure 2 ,like Figure 2 As shown, in this embodiment... Figure 1 Based on the examples, the product recommendation method is described in detail, which includes:

[0106] S201. Receive a product recommendation request. The product recommendation request includes: user identification information.

[0107] The explanation of step S201 is the same as that in the above embodiments, and will not be repeated here.

[0108] S202. Determine the intent recognition result corresponding to the user identification information.

[0109] The explanation of step S202 is the same as that in the above embodiments, and will not be repeated here.

[0110] S203. Through the first intelligent agent, perform anomaly assessment on the user profile associated with the user identification information and generate anomaly level.

[0111] The user profile includes, but is not limited to: occupation, age, assets, customer rating, investment risk preference, investment products held, and historical investment browsing history.

[0112] Anomaly level is used to characterize a user's level of risk tolerance. For example, if the anomaly level is "no anomaly," it indicates that the user's level of risk tolerance is higher; conversely, if the anomaly level is "high anomaly," it indicates that the user's level of risk tolerance is lower.

[0113] The purpose of this step is to verify the user's risk tolerance.

[0114] Understandably, user profiles associated with user identification information contain core data such as investment preferences and financial status, which can accurately represent a user's risk tolerance. If anomalies in this user profile are not assessed, it may lead to the recommendation of products that exceed the user's risk tolerance or do not match the user's actual situation.

[0115] Therefore, it is necessary to use a first intelligent agent to conduct anomaly assessment on the user profile associated with user identification information in order to determine the user's risk tolerance.

[0116] Optionally, this application provides a method for determining a user profile, including:

[0117] The first step is to use a fourth intelligent agent to interact with the new user if the user identification information indicates that the user is a new user. The agent will then conduct a question-and-answer interaction with the new user to obtain the question-and-answer results and determine the user profile of the new user based on the results.

[0118] The purpose of this step is to build user profiles for new users who lack historical user profile data, thus filling the information gaps for new users.

[0119] Understandably, when the user identification information indicates that the user is a new user, it means that the system does not yet have the user's historical transaction records; and the fourth intelligent agent has professional question-and-answer interaction capabilities, which can guide new users to actively provide key information in an organized and targeted manner.

[0120] Therefore, by engaging in question-and-answer interactions with new users through a fourth intelligent agent, core financial information such as users' investment preferences, financial status, and investment horizon can be collected in a targeted manner. These scattered question-and-answer results can be transformed into structured user profile data, providing a reliable basis for the subsequent recommendation process.

[0121] The second step is to obtain historical transaction data based on the user identification information if the user is a previous user, and to analyze and process the historical transaction data to obtain a user profile of the previous user.

[0122] Historical transaction data includes, but is not limited to, key information such as the type of financial products purchased by the user, the amount of each transaction, the holding period of the product, the transaction time, and the return feedback.

[0123] The purpose of this step is to build user profiles using historical transaction data from existing users.

[0124] Understandably, when user identification information indicates that a user is an old user, it means that the user has a history of using the system for a period of time and has already made transactions. This transaction data can represent the user's true financial preferences.

[0125] Therefore, by obtaining a user's historical transaction data through user identification information and analyzing and processing this data, we can update and improve the user profile that matches the old user, providing reliable data support for the subsequent recommendation process.

[0126] S204. Using a second intelligent agent, multiple recommended products are determined based on the anomaly level and intent recognition results.

[0127] The purpose of this step is to recommend multiple products that suit the user's needs and risk tolerance by analyzing the user's anomaly level and intent recognition results.

[0128] Understandably, firstly, the anomaly level can characterize a user's risk tolerance. Secondly, the intent recognition result can characterize a user's needs and preferences. Finally, the second agent can dynamically integrate the anomaly level and intent recognition result. On the one hand, it can adjust the strictness of the recommendation strategy based on the anomaly level, for example, prioritizing low-risk products for users with high anomalies. On the other hand, it can combine the intent recognition result to filter product types that match the user's goals.

[0129] Therefore, by utilizing a second intelligent agent to comprehensively consider the anomaly level and intent recognition results, multiple products that are more suitable for users and align with their investment goals and risk tolerance can be recommended.

[0130] Optionally, this application provides a possible implementation method, including:

[0131] The first step is to obtain a product knowledge graph through a second intelligent agent. The product knowledge graph includes multiple candidate products and the product attributes corresponding to each candidate product.

[0132] Product attributes include, but are not limited to: product name, type, price, range, sales time range, sales volume, and profit margin.

[0133] Understandably, firstly, the second intelligent agent has product screening capabilities, and directly obtaining the product knowledge graph can avoid the redundancy of data transmission between multiple modules and improve screening efficiency.

[0134] Secondly, the product knowledge graph integrates candidate products and their attributes in a structured form, which makes it easier for the second intelligent agent to quickly retrieve and compare information compared to scattered data, thus ensuring the accuracy of the screening.

[0135] Therefore, acquiring product knowledge graphs through a second intelligent agent can provide complete and accurate basic data for subsequent product recommendations and analysis.

[0136] The second step is to determine the product level of each candidate product based on its corresponding product attributes.

[0137] Product grades include, but are not limited to, low grade and high grade. A higher grade product means that the product has lower risk and higher reliability, while a lower grade product means that the product has higher risk.

[0138] Understandably, different products carry different risks. Therefore, by comprehensively considering the attributes of each candidate product, the risk level of each product can be assessed, thus providing a basis for subsequent decision-making.

[0139] The third step is to determine multiple recommended products from multiple candidate products based on the intent recognition results, the anomaly level, and the product level corresponding to each candidate product.

[0140] The purpose of this step is to identify multiple products that suit the user's needs and risk tolerance.

[0141] Understandably, firstly, intent recognition results can clarify user needs, such as product type, investment period, and return goals.

[0142] Secondly, the anomaly level can characterize a user's risk tolerance. Ignoring this level may result in recommending products that exceed the user's actual risk tolerance, leading to compatibility risks.

[0143] Finally, product rating can characterize the risk level of a product. Higher-rated products have lower risks and higher reliability, while lower-rated products have higher risks.

[0144] Therefore, by comprehensively considering the intent recognition results, the anomaly level, and the product level corresponding to each candidate product, we can not only ensure that the recommended products meet the user's needs, but also avoid the risk of incompatibility issues, and finally select multiple recommended products with strong adaptability and high security.

[0145] S205. Using a third intelligent agent, generate recommendation reasons for each recommended product based on multiple recommended products, and determine the product recommendation result based on each recommended product and its corresponding recommendation reasons.

[0146] The purpose of this step is to generate a reason for recommending each product, and then combine these reasons to determine a reasonable product recommendation result.

[0147] Understandably, when choosing products, users typically want to understand the reasons behind the recommendations, rather than simply seeing a list of products. Therefore, generating the reasons for each recommended product through a third-party intelligent agent can help users understand the logic behind the recommendations, thereby increasing their acceptance and trust in the recommendations.

[0148] Optionally, this application provides a possible implementation method, including: inputting the product attributes corresponding to each recommended product into a language generation model through a third intelligent agent to obtain the recommendation reasons corresponding to each recommended product.

[0149] Among these features, the language generation model possesses powerful natural language processing and logical integration capabilities, enabling it to deeply analyze the core attributes of each recommended product. Therefore, by inputting the attributes of each recommended product into the language generation model, a unique recommendation rationale can be generated for each product, helping users clearly understand how the product matches their needs and risk tolerance.

[0150] S206, Output product recommendation results.

[0151] The explanation of step S206 is the same as that in the above embodiments, and will not be repeated here.

[0152] S207. Obtain multimodal data.

[0153] Multimodal data is used to characterize users' real-time behavioral states. Multimodal data includes, but is not limited to: clicks / favorites / purchases of recommended products, voice consultation content, and page browsing history.

[0154] Understandably, after outputting product recommendation results, by acquiring users' multimodal data, we can analyze users' real feedback on the recommended content, thereby providing a basis for subsequent optimization of recommendation strategies.

[0155] S208. Input the multimodal data into the behavioral feature analysis model to obtain the behavioral feature results output by the behavioral feature analysis model.

[0156] Understandably, by analyzing user multimodal interaction data through behavioral feature analysis models, structured and actionable behavioral feature results can be extracted from scattered multimodal data. This can provide data support for subsequent updates to user profiles, adjustments to product selection strategies, and optimization of intelligent agent interaction logic.

[0157] S209. Based on the behavioral characteristics, adjust the product recommendation results to obtain the adjusted product recommendation results, and output the adjusted product recommendation results.

[0158] The purpose of this step is to achieve accurate matching of personalized recommendations and optimization of the user experience by dynamically adjusting the product recommendation results.

[0159] It is understandable that behavioral characteristics can represent a user's interests, habits and preferences. However, the initial product recommendation results are generated based solely on the initial user profile and intent recognition results, without fully considering the user's real-time interactive feedback and dynamic changes in needs after receiving the recommendation, which may result in adaptation bias.

[0160] Therefore, by using behavioral feature results as adjustment factors to filter, sort, or supplement the initial product recommendation results, the final output recommendation results can be more in line with the user's current actual state and real needs, thereby effectively improving user satisfaction.

[0161] The product recommendation method provided in this application first receives a product recommendation request containing user identification information and determines the corresponding intent recognition result based on the user identification information; then, a first intelligent agent performs anomaly assessment on the user profile associated with the user identification information and generates anomaly level; a second intelligent agent combines the anomaly level with the intent recognition result to filter out multiple recommended products; a third intelligent agent generates corresponding recommendation reasons for each recommended product, thereby determining the initial product recommendation result; simultaneously, multimodal data is acquired and input into a behavioral feature analysis model to obtain behavioral feature results; the initial product recommendation result is adjusted and optimized based on the results; and finally, the adjusted product recommendation result is output.

[0162] This method, through multi-agent collaboration, combined with anomaly analysis of user profiles and multimodal behavioral feature mining, overcomes the limitations of traditional fixed rules and achieves precise matching of diverse and personalized user needs. Simultaneously, the anomaly evaluation stage improves the security and adaptability of the recommendation results, and the use of multimodal data makes the recommendations more aligned with users' actual behavioral preferences. Furthermore, the generation of recommendation reasons not only enhances the interpretability and credibility of the recommendation results but also effectively improves user satisfaction with product recommendations.

[0163] Figure 3 A schematic diagram of the product recommendation device provided in this application is shown below. Figure 3 As shown, the product recommendation device 300 provided in this embodiment includes:

[0164] The receiving module 301 is used to receive a product recommendation request, which includes user identification information.

[0165] The determination module 302 is used to determine the intent recognition result corresponding to the user identification information;

[0166] The determining module 302 is also used to determine the product recommendation result corresponding to the intent recognition result through the collaborative recommendation intelligent agent group. The product recommendation result includes: multiple recommended products and the recommendation reason corresponding to each recommended product.

[0167] Output module 303 is used to output product recommendation results.

[0168] In one possible implementation, the apparatus further includes: a generation module 304;

[0169] The generation module 304 is used to perform anomaly assessment on the user profile associated with the user identification information through the first intelligent agent and generate anomaly level;

[0170] The determination module 302 is specifically used to determine multiple recommended products based on the anomaly level and intent recognition results through the second intelligent agent;

[0171] The determination module 302 is specifically used to generate recommendation reasons for each recommended product based on multiple recommended products through a third intelligent agent, and to determine the product recommendation result based on each recommended product and its corresponding recommendation reasons.

[0172] In one possible implementation, if the user identification information indicates that the user is a new user, the determining module 302 is further configured to engage in question-and-answer interaction with the new user through a fourth intelligent agent, obtain the question-and-answer results, and determine the user profile of the new user based on the question-and-answer results.

[0173] The device also includes: an acquisition module 305;

[0174] The acquisition module 305 is used to acquire historical transaction data based on the user identification information if the user identification information indicates that the user is an old user.

[0175] The device also includes: a processing module 306;

[0176] Processing module 306 is used to analyze and process historical transaction data to obtain user profiles of old users.

[0177] In one possible implementation, the acquisition module 305 is further configured to acquire a product knowledge graph through a second intelligent agent. The product knowledge graph includes multiple candidate products and product attributes corresponding to each candidate product.

[0178] The determination module 302 is also used to determine the product level of each candidate product based on the product attributes corresponding to each candidate product.

[0179] The determination module 302 is specifically used to determine multiple recommended products from multiple candidate products based on the intent recognition results, the anomaly level, and the product level corresponding to each candidate product.

[0180] In one possible implementation, the device further includes: an input module 307;

[0181] The input module 307 is used to input the product attributes corresponding to each recommended product into the language generation model through a third intelligent agent to obtain the recommendation reasons for each recommended product.

[0182] In one possible implementation, the acquisition module 305 is also used to acquire a preset filter word list through a fourth intelligent agent during the question-and-answer interaction process.

[0183] The receiving module 301 is also used to receive questions from the first user;

[0184] Processing module 306 is also used to filter the first user question based on the filter vocabulary to obtain the second user question;

[0185] The determination module 302 is specifically used to conduct question-and-answer interaction with the user based on the second user's question until the question-and-answer result is obtained.

[0186] In one possible implementation, the acquisition module 305 is further configured to acquire multimodal data;

[0187] The input module 307 is also used to input multimodal data into the behavior feature analysis model to obtain the behavior feature results output by the behavior feature analysis model;

[0188] The processing module 306 is also used to adjust the product recommendation results based on the behavioral feature results, obtain the adjusted product recommendation results, and output the adjusted product recommendation results.

[0189] The product recommendation device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0190] Figure 4 A structural diagram of the recommended equipment for the product provided in this application. (Example) Figure 4 As shown, the electronic device 400 provided in this embodiment includes at least one processor 401 and a memory 402. Optionally, the device 400 further includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus 404.

[0191] In a specific implementation, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to perform the above-described method.

[0192] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0193] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0194] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0195] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0196] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0197] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0198] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0199] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0200] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0201] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0202] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0203] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0204] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0205] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A product recommendation method, characterized in that, The method includes: Receive a product recommendation request, the product recommendation request including: user identification information; Determine the intent recognition result corresponding to the user identification information; By using a collaborative recommendation intelligent agent group, a product recommendation result corresponding to the intent recognition result is determined. The product recommendation result includes: multiple recommended products and the recommendation reason for each recommended product. Output the product recommendation results.

2. The method according to claim 1, characterized in that, The collaborative recommendation agent group includes: a first agent, a second agent, and a third agent. The step of determining the product recommendation result corresponding to the intent recognition result through the collaborative recommendation agent group includes: The first intelligent agent performs anomaly assessment on the user profile associated with the user identification information and generates anomaly level. The second intelligent agent determines the multiple recommended products based on the anomaly level and the intent recognition result. The third intelligent agent generates recommendation reasons for each of the multiple recommended products, and determines the product recommendation result based on each recommended product and its corresponding recommendation reasons.

3. The method according to claim 2, characterized in that, The collaborative recommendation agent group further includes: a fourth agent, and the method further includes: If the user identification information indicates that the user is a new user, then the fourth intelligent agent will conduct a question-and-answer interaction with the new user to obtain the question-and-answer results, and based on the question-and-answer results, determine the user profile of the new user; If the user identification information indicates that the user is a previous user, then based on the user identification information, historical transaction data is obtained, and the historical transaction data is analyzed and processed to obtain a user profile of the previous user.

4. The method according to claim 2, characterized in that, The step of determining the multiple recommended products through the second intelligent agent based on the anomaly level and the intent recognition result includes: The second intelligent agent obtains a product knowledge graph, which includes multiple candidate products and product attributes corresponding to each candidate product. Based on the product attributes corresponding to each of the candidate products, the product level corresponding to each of the candidate products is determined; Based on the intent recognition result, the anomaly level, and the product level corresponding to each candidate product, the multiple recommended products are determined from the multiple candidate products.

5. The method according to claim 4, characterized in that, The step of generating recommendation reasons for each of the multiple recommended products through the third intelligent agent includes: The third intelligent agent inputs the product attributes corresponding to each recommended product into the language generation model to obtain the recommendation reasons for each recommended product.

6. The method according to claim 3, characterized in that, The step of interacting with the new user through the fourth intelligent agent to obtain the question-and-answer results includes: During the question-and-answer interaction, the fourth intelligent agent obtains a preset filter word list; Receive a first user question, and filter the first user question based on the filter vocabulary to obtain a second user question; Based on the second user's question, a question-and-answer interaction is conducted with the user until the question-and-answer result is obtained.

7. The method according to claim 1, characterized in that, The method further includes: Acquire multimodal data; The multimodal data is input into the behavior feature analysis model to obtain the behavior feature results output by the behavior feature analysis model; Based on the behavioral characteristics, the product recommendation results are adjusted to obtain the adjusted product recommendation results, and the adjusted product recommendation results are output.

8. A product recommendation device, characterized in that, include: The receiving module is used to receive product recommendation requests, wherein the product recommendation requests include: user identification information; The determination module is used to determine the intent recognition result corresponding to the user identification information; The determining module is further configured to determine the product recommendation result corresponding to the intent recognition result through a collaborative recommendation intelligent agent group, wherein the product recommendation result includes: multiple recommended products and the recommendation reason corresponding to each recommended product; The output module is also used to output the product recommendation results.

9. A product recommendation device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

11. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images