Financial service recommendation method and apparatus
By combining multi-path recall and large language models, the problems of insufficient accuracy and weak interpretability in financial business recommendations have been solved, achieving efficient and personalized financial business recommendations and improving user trust and service efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- IND BANK CO
- Filing Date
- 2026-03-19
- Publication Date
- 2026-05-29
AI Technical Summary
Financial service recommendations lack in-depth user insights, resulting in insufficient accuracy, weak interpretability of recommendation results, low user trust, difficulty in covering massive long-tail questions by traditional question-and-answer systems, and low customer service response efficiency.
By combining a multi-path recall method with a large language model, user interest tags are determined, financial business information is refined and rearranged, explainable recommendation reasons are provided, and RAG technology is introduced to answer user questions, combined with financial service documents.
It improved the accuracy of financial service recommendations, matched users' real needs, enhanced user trust and service efficiency, and strengthened personalized experience and professional credibility.
Smart Images
Figure CN122115079A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a method and apparatus for recommending financial services. Background Technology
[0002] In the banking industry, precise recommendations are a crucial means to improve business conversion rates and customer satisfaction. However, traditional product recommendations in financial scenarios suffer from the following significant pain points: Product recommendations lack deep user insight and fail to adequately integrate users' current product preferences with their actual asset situation, resulting in insufficient accuracy and difficulty in matching users' true needs. The interpretability of the recommendation results is weak; when users have questions about the recommendation logic, they lack convenient and clear ways to obtain the underlying evidence, reducing user trust.
[0003] This section is intended to provide background or context for the embodiments of this application set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section. Summary of the Invention
[0004] To address at least one problem in the prior art, this application proposes a financial service recommendation method and apparatus that can improve the accuracy of financial service recommendations and match users' real needs.
[0005] To address the aforementioned technical problems, this application provides the following technical solution: Firstly, this application provides a method for recommending financial services, including: Acquire financial business information sets, target users' user attributes, and historical behavior data; Based on user attributes and historical behavior data, determine the interest tags of the target user; Based on the financial business information set, the user attributes and historical behavior data of the target users, a multi-path recall method is used to determine the preliminary financial business information to be recommended. The multi-path recall method includes: a recall method based on the interest tags, a collaborative filtering recall method, and a model recall method. Based on the historical behavior data, the preliminary financial business information to be recommended is refined and rearranged to obtain the final financial business information to be recommended and output.
[0006] In one embodiment, the step of refining and rearranging the initial financial business information to be recommended based on the historical behavior data to obtain and output the final financial business information to be recommended includes: Based on historical behavioral data, a multi-objective model is used to refine the preliminary financial business information to be recommended, thereby obtaining intermediate financial business information to be recommended. Based on the historical behavior data, the intermediate financial business information to be recommended is rearranged using a large language model to obtain the final financial business information to be recommended, and the final financial business information to be recommended and the pre-acquired recommendation reasons are output.
[0007] In one embodiment, before determining the initial financial business information to be recommended based on the financial business information set, the target user's user attributes, and historical behavior data through a multi-channel recall method, the method further includes: The interest tags of the target user are generalized to obtain a generalization result; correspondingly, the recall method based on the interest tags is a recall method based on the generalization result.
[0008] In one embodiment, the financial service recommendation method further includes: Receive questions from the target user regarding the final financial service information to be recommended; Obtain the target user's historical dialogue records and historical recommendation results, and generate the answer to the question based on the question, the target user's historical dialogue records, historical recommendation results, and the final financial business information to be recommended; Determine whether the answer to the question meets the preset answer requirements. If so, output the answer to the question.
[0009] In one embodiment, the financial service recommendation method further includes: Obtain the current click behavior of the target user and apply the current click behavior to update the target user's historical behavior data.
[0010] In one embodiment, determining the target user's interest tags based on user attributes and historical behavior data includes: Based on user attributes and historical behavior data, the interest tags of the target user are determined through interaction depth analysis. The interest tags include long-term interest tags and short-term interest tags.
[0011] Secondly, this application provides a financial service recommendation device, comprising: The acquisition module is used to acquire financial business information sets, user attributes of target users, and historical behavior data; The determination module is used to determine the interest tags of the target user based on user attributes and historical behavior data; The recall module is used to determine the initial financial business information to be recommended based on the financial business information set, the user attributes and historical behavior data of the target user, and through a multi-path recall method. The multi-path recall method includes: a recall method based on the interest tags, a collaborative filtering recall method, and a model recall method. The recommendation module is used to refine and rearrange the preliminary financial business information to be recommended based on the historical behavior data, and to obtain and output the final financial business information to be recommended.
[0012] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the financial business recommendation method.
[0013] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the financial business recommendation method.
[0014] Fifthly, this application provides a computer program product, which includes a computer program that, when executed by a processor, implements the financial business recommendation method.
[0015] As can be seen from the above technical solution, this application provides a method and apparatus for recommending financial services. The method includes: acquiring a set of financial service information, user attributes of a target user, and historical behavior data; determining the target user's interest tags based on the user attributes and historical behavior data; determining preliminary financial service information to be recommended based on the financial service information set, user attributes of the target user, and historical behavior data through a multi-path recall method, wherein the multi-path recall method includes: a recall method based on the interest tags, a collaborative filtering recall method, and a model recall method; and re-ranking and rearranging the preliminary financial service information to be recommended based on the historical behavior data to obtain and output the final financial service information to be recommended. This improves the accuracy of financial service recommendations and matches users' real needs. Specifically, it enhances the personalized customer experience by recommending financial services that users are interested in, based on the world knowledge of a large model, providing reasons for recommendations, and enabling interaction based on the large model to answer user questions, while also providing answers in conjunction with professional financial service documents and relevant reference information, thus improving service efficiency and professional credibility. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a first flowchart illustrating the financial business recommendation method in the embodiments of this application; Figure 2 This is a second flowchart illustrating the financial business recommendation method in the embodiments of this application; Figure 3 This is a second flowchart illustrating the financial business recommendation method in the embodiments of this application; Figure 4 This is a schematic diagram of the financial business recommendation device in the embodiments of this application; Figure 5 This is a logical schematic diagram of the financial business recommendation device in the application example of this application; Figure 6 This is a logical diagram of the user interest extraction module in the application example of this application; Figure 7 This is a logical diagram of the product filtering and sorting module in the application example of this application; Figure 8 This is a logical diagram of the product recommendation module in the application example of this application; Figure 9 This is a logical diagram of the product consultation and Q&A module in the application example of this application; Figure 10 This is a schematic block diagram of the system configuration of an electronic device according to an embodiment of this application. Detailed Implementation
[0017] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0018] Existing technologies still have some problems: the interpretability of recommendation results is weak, and when users have questions about the recommendation logic, there is a lack of convenient and clear ways to obtain the underlying evidence, which reduces user trust; in the user consultation process, customer service response efficiency is low, and traditional Q&A systems are unable to cover a large number of long-tail questions, and cannot respond to users' personalized questions in a timely manner, affecting service experience and conversion results.
[0019] To address at least one of the problems existing in the prior art, this application proposes a financial business recommendation method and apparatus for financial scenarios. This method provides intelligent recommendations of financial products and services, enhancing the personalized customer experience. It recommends financial services of interest to users based on their preferences, providing reasons for the recommendations. Based on a large model, it enables interaction, answering user questions and incorporating professional financial service documentation to provide relevant reference information, thus improving service efficiency and enhancing professional credibility. Specifically, building upon the recall and ranking of traditional recommendation engines, it proposes a closed-loop recommendation process: preference extraction based on user attributes and historical behavior, combined with reasonable generalization based on world knowledge; recommending the user's most interesting product set based on a multi-path recall and fine-ranking model; re-ranking the recommendation results and writing recommendation reasons based on current context information, providing an interpretable recommendation result; and providing professional answers to user questions about the recommendation results, employing RAG technology to reduce answer illusions and performing secondary verification, controlling responses based on thresholds, and directly transferring unreliable responses to human intervention.
[0020] The financial business recommendation method and apparatus disclosed in this application can be used in the field of financial technology, as well as in any other field. The application areas of the financial business recommendation method and apparatus disclosed in this application are not limited. The acquisition, storage, use, and processing of data in the technical solutions of each embodiment of this application all comply with relevant laws and regulations.
[0021] The information collected in the technical solution of this application is information and data authorized by the user or fully authorized by all parties. 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, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.
[0022] The acquisition, transmission, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.
[0023] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.
[0024] The following examples illustrate this in detail.
[0025] To improve the accuracy of financial service recommendations and match users' actual needs, this embodiment provides a financial service recommendation method whose execution entity is a financial service recommendation device. This financial service recommendation device includes, but is not limited to, a server, such as... Figure 1 As shown, this method specifically includes the following: Step 100: Obtain financial business information set, target user attributes, and historical behavior data.
[0026] Specifically, the financial business information set may include: batch financial business information, each of which may include: an introduction to the corresponding financial business (e.g., product introduction, product manual), market purchasing trends, and purchasing policies; the user attributes may include: user age, gender, and occupation, etc.; historical behavioral data may include: click behavior, browsing time, browsing history, and final transaction records, etc. The financial business information set, target user attributes, and historical behavioral data can be obtained from a local database. The financial businesses may include: financial products and services; for example, demand deposits, time deposits, large-denomination certificates of deposit, insurance services, deposit and loan services, payment and settlement, etc.
[0027] Step 200: Determine the interest tags of the target user based on user attributes and historical behavior data.
[0028] For example, suppose the tags include: stocks, funds, time deposits, demand deposits, precious metals, etc.; based on user attributes and historical behavior data, the target user's interest tags, such as time deposits and precious metals, can be selected from the tags.
[0029] Step 300: Based on the financial business information set, the user attributes and historical behavior data of the target user, determine the preliminary financial business information to be recommended through a multi-path recall method. The multi-path recall method includes: a recall method based on the interest tags, a collaborative filtering recall method, and a model recall method.
[0030] Specifically, based on the target user's user attributes and historical behavior data, financial business information corresponding to each recall method can be selected from the financial business information set, and all of these can be used as the original financial business information to be recommended. A preliminary score can be determined based on the preset correspondence between recall methods and scores, and the recall methods corresponding to each original financial business information to be recommended. If multiple recall methods correspond to the same original financial business information to be recommended, the scores of each recall method corresponding to that initial financial business information to be recommended can be accumulated to determine the preliminary score of that original financial business information to be recommended. All original financial business information to be recommended with a preliminary score greater than the preliminary score threshold is determined as initial financial business information to be recommended. The preset correspondence between recall methods and scores can be set according to actual circumstances, and this application does not impose any restrictions on this.
[0031] Step 400: Perform fine sorting and rearrangement on the preliminary financial business information to be recommended, and output the final financial business information to be recommended.
[0032] Specifically, the preliminary financial business information to be recommended can be fine-ranked based on the historical behavior data, and the fine-ranking results can be rearranged based on the historical behavior data to obtain the final financial business information to be recommended and output for display.
[0033] To improve the interpretability of recommendation results, it is essential to have clear methods for obtaining the underlying evidence and to enhance user trust, such as... Figure 2 As shown, in one embodiment, step 400 includes: Step 401: Based on historical behavior data, a multi-objective model is used to refine the preliminary financial business information to be recommended, thereby obtaining intermediate financial business information to be recommended.
[0034] Specifically, historical behavioral data and each preliminary financial business information to be recommended can be input into a multi-objective model to obtain the fine ranking score corresponding to the preliminary financial business information to be recommended. The preliminary financial business information to be recommended with a fine ranking score greater than the fine ranking score threshold is identified as intermediate financial business information to be recommended.
[0035] Specifically, a refined ranking model can be used, which is trained based on users' historical click behavior and aims to maximize user conversion. To improve the robustness of the recommendation effect, a multi-objective model is adopted here, which, in addition to the click objective, includes browsing time and final transaction as objectives. Models such as ESMM and MMOE make the recommendation results more accurate and stable compared to single-objective models.
[0036] Step 402: Based on the historical behavior data, the intermediate financial business information to be recommended is rearranged using a large language model to obtain the final financial business information to be recommended, and the final financial business information to be recommended and the pre-acquired recommendation reasons are output.
[0037] Specifically, historical behavioral data and each intermediate financial transaction to be recommended can be input into a large language model to obtain a final score for that intermediate financial transaction. Intermediate financial transactions with a final score greater than a threshold are then identified as final financial transactions to be recommended. The pre-acquired recommendation reasons can be preset according to actual circumstances, such as "You have recently browsed this frequently," or "Recommended for you based on your investment risk aversion preferences."
[0038] Specifically, step 401 yields a preliminary personalized recommendation result. However, users, relying solely on the limited title information of the recommendation list, find it difficult to quickly grasp the displayed information, leading to a bias in their perception of the recommendation effect. Therefore, a personalized recommendation result reordering method is proposed. This method leverages the reasoning capabilities of a large model to reorder the recommendation results based on process information (such as browsing history, user preferences, product descriptions, and market dynamics). Furthermore, the large model can be used to summarize customer behavior and product information to generate the recommendation reasons, significantly improving the interpretability of the recommendation results. Additionally, the large model can be used to summarize and reorder the results based on customer behavior and product information to obtain and output the final financial business information to be recommended.
[0039] To avoid insufficient recommendation richness due to insufficient behavior, in one embodiment, before step 300, the method further includes: The interest tags of the target user are generalized to obtain a generalization result; correspondingly, the recall method based on the interest tags is a recall method based on the generalization result.
[0040] Specifically, the interest tags of the target users can be generalized, and generalization results are obtained based on the world knowledge of the large model. After tagging, the model's world knowledge is used to generalize the tags. For example, if a user is interested in information related to gold investment, then they are interested in safe-haven investments and inflation-hedging plans, and information with relevant tags can be recommended. This module greatly improves the problem of insufficient recommendation richness caused by insufficient behavioral data in traditional recommendation models. After obtaining the generalization results, all product content to be recommended can be recalled. This employs multi-path recall, such as model-based recall, collaborative filtering recall, and most importantly, recall based on user preferences. User preferences integrate the results of the large model generalization, improving the richness of the recommendation results. Figure 3 As shown.
[0041] To improve the accuracy of financial service recommendations while enhancing the user experience during subsequent consultations, such as... Figure 3 As shown, in one embodiment, after step 400 or step 402, the method further includes: Step 500: Receive questions from the target user regarding the final financial service information to be recommended.
[0042] Step 600: Obtain the target user's historical dialogue records and historical recommendation results, and generate the answer to the question based on the question, the target user's historical dialogue records, historical recommendation results, and the final financial business information to be recommended.
[0043] Specifically, the historical recommendation results may include: user attitude, historically successfully recommended financial business information, number of successful recommendations, and number of failed recommendations. Keywords can be extracted from the question, the target user's historical dialogue records, historical recommendation results, and the final financial business information to be recommended. Based on these keywords, a matching answer can be determined from a pre-defined answer database as the answer to the question. Alternatively, an answer to the question can be generated based on a large model + RAG (Research and Development Authority) framework, using the question, the target user's historical dialogue records, historical recommendation results, and the final financial business information to be recommended.
[0044] Step 700: Determine whether the answer to the question meets the preset answer requirements. If so, output the answer to the question.
[0045] Specifically, a large-scale model-based answer evaluation module can be used to determine whether the answer to the question meets the preset answer requirements. If so, the answer to the question is output. The preset answer requirements can be set according to the actual situation, and this application does not impose any restrictions on this. After showing the recommended results to the user, if the user is interested, they will often inquire about the product content. Traditionally, a large amount of document information is displayed for the user to view. To improve the user's search efficiency, this application provides a question-and-answer module. First, in the answer retrieval stage, not only is the dialogue history combined, but product information, purchase policies, and recommended results are also put into the memory management module (e.g., a database) as context to enhance the reliability of the answer. At the same time, the source of the answer is also attached to ensure that the content is traceable. To improve the reliability of the answer and the user experience, a security mechanism is introduced to perform secondary verification of the results. It can analyze the answer content based on dimensions such as compliance, accuracy, degree of illusion, and degree of resolution. If any one of them is not met, it will be transferred to human intervention to ensure service quality. If all of them are met, it can be determined that the answer to the question meets the preset answer requirements. Users often provide further feedback on the results, such as clicking or not clicking. These behaviors are further saved to form a data loop. The recommendation model is also trained based on the feedback behavior to continuously improve the efficiency of recommendation and shopping guidance.
[0046] In one embodiment, the financial service recommendation method further includes: obtaining the current click behavior of the target user and applying the current click behavior to update the target user's historical behavior data.
[0047] Specifically, the click behavior can be added to the target user's historical behavior data in the local database to update the target user's historical behavior data, which will facilitate the next financial service recommendation for the target user.
[0048] To improve the reliability of determining interest tags, in one embodiment, step 200 includes: Based on user attributes and historical behavior data, the interest tags of the target user are determined through interaction depth analysis. The interest tags include long-term interest tags and short-term interest tags.
[0049] Specifically, by gaining in-depth insights into users, adjustments can be made according to the actual scenario. There are two key steps: First, preference tags are extracted based on user attributes and historical behavior data. It is necessary to tag behaviors and products. Second, users are tagged in reverse through interaction depth, taking into account long-term and short-term interests, and interest tags can be created separately.
[0050] From a software perspective, in order to improve the accuracy of financial service recommendations and match users' actual needs, this application provides an embodiment of a financial service recommendation device for implementing all or part of the aforementioned financial service recommendation method. See [link to embodiment]. Figure 4 The financial service recommendation device specifically includes the following components: Module 01 is used to acquire financial business information sets, user attributes of target users, and historical behavior data. Module 02 is used to determine the interest tags of the target user based on user attributes and historical behavior data; The recall module 03 is used to determine the initial financial business information to be recommended based on the financial business information set, the user attributes of the target user and historical behavior data, through a multi-path recall method. The multi-path recall method includes: a recall method based on the interest tags, a collaborative filtering recall method and a model recall method. Recommendation module 04 is used to refine and rearrange the preliminary financial business information to be recommended based on the historical behavior data, and to obtain and output the final financial business information to be recommended.
[0051] In one embodiment, the financial service recommendation device further includes: The product consultation module is used to generate answers to the questions based on a large model + RAG. The answer evaluation module based on the large model is used to determine whether the answers to the questions meet the preset answer requirements. If so, the answer to the questions is output.
[0052] The embodiments of the financial business recommendation device provided in this specification can be used to execute the processing flow of the embodiments of the above-described financial business recommendation method. Its functions will not be repeated here, but can be referred to the detailed description of the embodiments of the above-described financial business recommendation method.
[0053] To further illustrate this solution, this application provides an application example of a financial business recommendation device. In this application example, such as... Figure 5As shown, the device includes: a user interest extraction module, a product filtering and sorting module, a product recommendation module, and a product consultation and Q&A module. The user interest extraction module can be considered equivalent to the combined functions of the acquisition module and the determination module. The combined functions of the product filtering and sorting module and the product recommendation module can be considered equivalent to the combined functions of the recall module and the recommendation module. A detailed description follows: The user interest extraction module provides in-depth insights into users and can be adjusted according to the actual scenario. It involves two key steps, such as... Figure 6 As shown, the process first extracts semantic tags for user preferences based on user attributes and historical behavioral data (e.g., browsing behavior), and then generalizes the semantics of the large model. Specifically, behaviors and products need to be tagged, and users are tagged in reverse based on interaction depth, considering both long-term and short-term interests, which can be tagged separately. After tagging, the tags are generalized using the world knowledge of the large model. For example, if a user is interested in information related to gold investment, then they are also interested in hedging investments and inflation-hedging planning, and information with relevant tags can be recommended. This can greatly improve the problem of insufficient recommendation richness caused by insufficient behavioral data in traditional recommendation models.
[0054] Product filtering and sorting module, such as Figure 7 As shown, this module first retrieves a candidate set of user interests (i.e., all product content to be recommended). This employs multi-path retrieval, such as model-based retrieval, collaborative filtering retrieval, and most importantly, content-based retrieval (i.e., retrieval based on user preferences). User preferences are integrated with the generalization results of the aforementioned large models, enhancing the richness of the recommendation results. Next, a ranking model (i.e., a fine-grained ranking model) is trained. This model can be constructed based on user behavior characteristics, building training samples for the fine-grained ranking model. For example, the fine-grained ranking model can be trained based on users' historical click behavior, aiming to maximize user conversion. To improve the robustness of the recommendation effect, a multi-objective model is used. In addition to the click objective, browsing time and final transaction are also included as objectives, such as ESMM and MMOE models, making the recommendation results more accurate and stable compared to single-objective models.
[0055] The product recommendation module, after fine-tuning, yields a rudimentary form of personalized recommendations. However, users, relying solely on the limited information in the recommendation list's titles, find it difficult to quickly grasp the displayed information, leading to a biased perception of the recommendation's effectiveness. This application example proposes a personalized reordering of the recommendation results, such as... Figure 8As shown, the product recommendation module utilizes the reasoning capabilities of a large model to rearrange the data based on process information (such as browsing history, user preferences, product descriptions, market dynamics, etc.) using an LLM large language model, generating recommendation reasons and displaying the recommendation results. In other words, the recommendation results obtained after fine-tuning can be rearranged a second time based on process information, while generating recommendation reasons (such as your recent frequent browsing or recommendations based on your investment risk aversion preferences), which can significantly improve the interpretability of the recommendation results.
[0056] The product consultation and Q&A module displays recommended results to users. If users are interested, they often inquire about the product content. Traditionally, this would involve displaying a large amount of documentation for users to view. This application example provides a product consultation and Q&A module to improve user search efficiency. This module can include a memory management module and an answer evaluation module. Figure 9 As shown, the system acquires user questions about products / services, characterizes the questions, and retrieves answers by combining product manuals, purchase policies, Q&A history, and recommendation results from the memory module. An answer is then generated, and the answer evaluation module analyzes the answer content for compliance, accuracy, perceptual appeal, and resolution, returning the answer or transferring it to a human agent. Specifically, the answer retrieval stage utilizes a comprehensive dialogue memory management module. This module not only incorporates dialogue history but also integrates product information, purchase policies, and recommendation results as context, enhancing the reliability of the answer. The answer also includes its source information to ensure traceability. To further improve answer reliability and user experience, a security mechanism is introduced for secondary verification of the results. The answer evaluation module analyzes the answer content for compliance, accuracy, perceptual appeal, and resolution; if any aspect is not met, the answer is transferred to a human agent to ensure service quality. Users often provide further feedback on the results, such as clicking or not clicking. These actions are then fed back to the user interest extraction module, forming a data loop. The recommendation model is also trained based on this feedback, continuously improving recommendation and shopping guidance efficiency.
[0057] As described above, the financial service recommendation method and apparatus provided in this application can achieve replicable recommendation and sales agent services, with clear modularity. Through continuous data feedback, it forms a self-iterative mechanism, effectively improving user experience. Based on interpretable information from the recommendation results, it can provide convenient consultation and Q&A, enhancing customer experience while reducing operational personnel costs. It can provide personalized recommendation results, increasing the stickiness of financial services and products, and improving business operation efficiency. Application scenarios include, but are not limited to, retail, corporate finance, and interbank businesses. The implementation of this application will significantly enhance the market penetration and user stickiness of financial services, reduce operating costs, and enhance customer experience, bringing higher business value and market competitiveness to financial institutions.
[0058] Figure 10 This is a schematic diagram of the physical structure of the electronic device provided in the embodiments of this application, such as... Figure 10 As shown, the electronic device includes: a memory 1001, a processor 1002, and a computer program stored in the memory 1001 and executable on the processor 1002. When the processor 1002 executes the computer program, it implements the following method: Acquire financial business information sets, target users' user attributes, and historical behavior data; Based on user attributes and historical behavior data, determine the interest tags of the target user; Based on the financial business information set, the user attributes and historical behavior data of the target users, a multi-path recall method is used to determine the preliminary financial business information to be recommended. The multi-path recall method includes: a recall method based on the interest tags, a collaborative filtering recall method, and a model recall method. Based on the historical behavior data, the preliminary financial business information to be recommended is refined and rearranged to obtain the final financial business information to be recommended and output.
[0059] This embodiment discloses a computer program product, which includes a computer program that, when executed by a processor, implements the following method: Acquire financial business information sets, target users' user attributes, and historical behavior data; Based on user attributes and historical behavior data, determine the interest tags of the target user; Based on the financial business information set, the user attributes and historical behavior data of the target users, a multi-path recall method is used to determine the preliminary financial business information to be recommended. The multi-path recall method includes: a recall method based on the interest tags, a collaborative filtering recall method, and a model recall method. Based on the historical behavior data, the preliminary financial business information to be recommended is refined and rearranged to obtain the final financial business information to be recommended and output.
[0060] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the following method: Acquire financial business information sets, target users' user attributes, and historical behavior data; Based on user attributes and historical behavior data, determine the interest tags of the target user; Based on the financial business information set, the user attributes and historical behavior data of the target users, a multi-path recall method is used to determine the preliminary financial business information to be recommended. The multi-path recall method includes: a recall method based on the interest tags, a collaborative filtering recall method, and a model recall method. Based on the historical behavior data, the preliminary financial business information to be recommended is refined and rearranged to obtain the final financial business information to be recommended and output.
[0061] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0062] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0063] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0064] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0065] In the description of this specification, the references to terms such as "an embodiment," "a specific embodiment," "some embodiments," "for example," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0066] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for recommending financial services, characterized in that, include: Acquire financial business information sets, target users' user attributes, and historical behavior data; Based on user attributes and historical behavior data, determine the interest tags of the target user; Based on the financial business information set, the user attributes and historical behavior data of the target users, a multi-path recall method is used to determine the preliminary financial business information to be recommended. The multi-path recall method includes: a recall method based on the interest tags, a collaborative filtering recall method, and a model recall method. Based on the historical behavior data, the preliminary financial business information to be recommended is refined and rearranged to obtain the final financial business information to be recommended and output.
2. The financial business recommendation method according to claim 1, characterized in that, The step of refining and rearranging the initial financial business information to be recommended based on the historical behavior data to obtain and output the final financial business information to be recommended includes: Based on historical behavioral data, a multi-objective model is used to refine the preliminary financial business information to be recommended, thereby obtaining intermediate financial business information to be recommended. Based on the historical behavior data, the intermediate financial business information to be recommended is rearranged using a large language model to obtain the final financial business information to be recommended, and the final financial business information to be recommended and the pre-acquired recommendation reasons are output.
3. The financial business recommendation method according to claim 1, characterized in that, Before determining the initial financial business information to be recommended based on the financial business information set, the user attributes of the target user, and historical behavior data through a multi-channel recall method, the following steps are also included: The interest tags of the target user are generalized to obtain a generalization result; correspondingly, the recall method based on the interest tags is a recall method based on the generalization result.
4. The financial business recommendation method according to any one of claims 1 to 3, characterized in that, Also includes: Receive questions from the target user regarding the final financial service information to be recommended; Obtain the target user's historical dialogue records and historical recommendation results, and generate the answer to the question based on the question, the target user's historical dialogue records, historical recommendation results, and the final financial business information to be recommended; Determine whether the answer to the question meets the preset answer requirements. If so, output the answer to the question.
5. The financial business recommendation method according to claim 4, characterized in that, Also includes: Obtain the current click behavior of the target user and apply the current click behavior to update the target user's historical behavior data.
6. The financial business recommendation method according to claim 1, characterized in that, The step of determining the target user's interest tags based on user attributes and historical behavior data includes: Based on user attributes and historical behavior data, the interest tags of the target user are determined through interaction depth analysis. The interest tags include long-term interest tags and short-term interest tags.
7. A financial service recommendation device, characterized in that, include: The acquisition module is used to acquire financial business information sets, user attributes of target users, and historical behavior data; The determination module is used to determine the interest tags of the target user based on user attributes and historical behavior data; The recall module is used to determine the initial financial business information to be recommended based on the financial business information set, the user attributes and historical behavior data of the target user, and through a multi-path recall method. The multi-path recall method includes: a recall method based on the interest tags, a collaborative filtering recall method, and a model recall method. The recommendation module is used to refine and rearrange the preliminary financial business information to be recommended based on the historical behavior data, and to obtain and output the final financial business information to be recommended.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the financial business recommendation method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the financial business recommendation method according to any one of claims 1 to 6.
10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the financial business recommendation method according to any one of claims 1 to 6.