Intelligent financial product recommendation system and method

Abstract

The invention discloses an intelligent financial product recommendation system and method, through a user feature acquisition module and a user behavior analysis module, explicit data and implicit data of a user can be comprehensively collected, and two independent financial product recommendation lists are generated. And the list fusion module effectively integrates the data to generate a comprehensive recommendation list based on the multi-dimensional data of the user. The standard feature mapping module further accurately matches the user features with the product features, high individuation and accuracy of the recommendation result are ensured, the individuation requirement of the user is met, and in addition, a user behavior analysis module and a product recommendation value calculation module based on real-time data are adopted. Through real-time monitoring and analysis of user browsing behaviors, the system can dynamically adjust recommendation strategies and generate an optimized financial product recommendation list. The real-time response mechanism ensures that the recommendation result is highly matched with the current demand of the user and the market condition, and the user experience and the satisfaction degree are improved.

Description

Technical Field

[0001] The present invention belongs to the field of financial product recommendation, and in particular, relates to an intelligent financial product recommendation system and method thereof. Background Art

[0002] The intelligent financial product recommendation system is a recommendation system based on advanced data analysis, machine learning and artificial intelligence technologies. It is designed specifically for the financial industry and aims to provide users with personalized and accurate financial product recommendation services, helping users quickly find the most suitable products in the complex financial product market.

[0003] Existing recommendation systems have significant limitations in terms of personalized recommendations. Most systems rely mainly on explicit user data, such as age, gender, geographic location, and past purchase records, for recommendations. Although this method can provide a certain degree of personalized services, explicit data can only reflect the basic attributes and historical behaviors of users, and cannot deeply capture implicit data such as users' potential interests, emotional tendencies, and dynamic behavior patterns. These implicit data often contain users' real needs and interests. If these data are ignored, the results generated by the recommendation system may deviate from the actual needs of users and fail to meet users' expectations for personalization and accuracy.

[0004] In addition, traditional recommendation systems generally lack the ability to respond to real-time user behavior and market changes, and usually make static recommendations based on fixed rules or algorithms. This approach cannot adjust the recommendation strategy in a timely manner when user needs change rapidly or market conditions fluctuate, causing the recommendation results to lag behind the actual needs of users and changes in the market environment, showing low flexibility and timeliness. For example, a user may have a new interest in a specific type of product at a certain moment, or a new popular product may appear on the market, but the existing system often cannot capture these changes in time and adjust the recommended content accordingly. This lagging recommendation method not only affects the user experience, but may also lead to a decrease in user trust in the recommendation results, thereby reducing the system's use effect and user satisfaction. As users' requirements for personalization and real-time performance continue to increase, these shortcomings of traditional recommendation systems have become increasingly prominent.

[0005] In view of this, the present invention is proposed. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide an intelligent financial product recommendation system and method thereof, which solves the problems raised in the above-mentioned background technology.

[0007] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0008] An intelligent financial product recommendation system, comprising: a user feature acquisition module, the user feature acquisition module is used to generate a first financial product recommendation list according to facial image features of a to-be-recommended object;

[0009] A user behavior analysis module, the user behavior analysis module is used to generate a second financial product recommendation list according to the length of time that the recommended object browses the financial product details page;

[0010] A list fusion module, which is used to fuse the first financial product recommendation list and the second financial product recommendation list to obtain a financial product recommendation list based on user implicit data;

[0011] A standard feature mapping module, which is used to obtain a standard feature mapping pair based on a candidate user feature group and a candidate product feature group generated by a financial product recommendation graph, and to fuse the standard user features and the standard product features in the standard feature mapping pair to obtain a standard recommendation feature;

[0012] A product recommendation value calculation module, which is used to generate a standard recommendation feature group based on all standard recommendation features, and calculate the product recommendation value of the financial product based on the standard recommendation feature group;

[0013] The financial product recommendation module is used to recommend financial products whose product recommendation values ​​are greater than the recommendation threshold to target financial users.

[0014] Optionally, the step of generating the first financial product recommendation list according to the facial image features of the object to be recommended is:

[0015] Capturing a face image in real time through a camera, preprocessing the acquired face image, analyzing the preprocessed face image, and extracting a feature vector containing face features;

[0016] Matching the extracted feature vector with the existing user feature database to identify the identity of the object to be recommended, and assigning the obtained identity of the object to be recommended to a specific user category;

[0017] According to the facial image features and the feature matching results, a user profile of the recommended object is constructed, and based on the information in the user profile, financial products that best match the user profile are screened from the financial product library to generate a first financial product recommendation list;

[0018] The initially generated financial product recommendation list is optimized and sorted, and finally a sorted first financial product recommendation list is formed based on the priority in the user portrait, the return-risk matching of the financial products and the current market conditions.

[0019] Optionally, the step of generating the second financial product recommendation list according to the time the recommended object browses the financial product details page is:

[0020] Monitor and record the browsing behavior of the recommended objects on the financial product details page in real time and calculate the total stay time of the user on each financial product details page based on the collected behavior data. The expression is: Among them, T p is the total length of time a user stays on a financial product details page, n is the number of effective stays performed by the user on the page, is the starting timestamp of the i-th entry into the page, that is, the time when the user enters the page, is the end timestamp of the i-th exit page, that is, the time when the user leaves the page, δ i is the effectiveness coefficient of the ith behavior;

[0021] Statistical analysis is performed on the browsing time of users on different product detail pages to identify the financial products that users focus on and assign an interest score I to each browsed financial product based on the user's browsing time and behavior analysis results. p , whose expression is: Among them, I p : Interest score of financial product p, w b is the behavior weight, is the sum of the weighted total stay time of the user on all the financial product detail pages that the user has browsed, m represents the total number of financial products that the user has browsed, and T j represents the total length of time a user stays on the jth product page, w j represents the behavior weight of the jth product, C p is the number of interactions of users when browsing financial product p, max(C j ) is the maximum number of interactions that occur on all financial product pages viewed by the user.

[0022] Combined with the user's interest score, a user's product preference model is constructed, and based on the user's product preference model, a group of candidate products that best match the user's preferences are screened from the financial product library to form a preliminary second financial product recommendation list;

[0023] The screened candidate products are optimized and sorted, and finally an optimized second financial product recommendation list is generated.

[0024] Optionally, the step of merging the first financial product recommendation list and the second financial product recommendation list to obtain a financial product recommendation list based on user implicit data is:

[0025] Receiving financial product data in a first financial product recommendation list and a second financial product recommendation list;

[0026] The products in the first financial product recommendation list and the second financial product recommendation list are weighted according to the system preset rules; the system preset rules are a set of strategies and parameters set according to user needs, market conditions and business logic to guide the weight allocation and priority score calculation. Then, the weight allocation results of the same financial product from the two lists are weighted and summed to calculate the comprehensive priority score of each product. For the same product appearing in both lists, its comprehensive score will be improved, and for the product appearing in only one list, the comprehensive score will be determined based on the weight of the list in which it is located;

[0027] When the same financial products exist in the two lists, the ranking of the products in the merged list is determined according to the priority score;

[0028] The financial products that have been through priority score calculation and conflict resolution are merged to form a unified financial product recommendation list. The merged list will include the financial products in the two recommendation lists and will be sorted by priority score;

[0029] Based on the user's personalized preferences and historical behavior data, the integrated financial product recommendation list is further optimized, and finally a financial product recommendation list based on the user's implicit data is generated.

[0030] Optionally, the steps of obtaining a standard feature mapping pair according to the candidate user feature group and the candidate product feature group generated by the financial product recommendation graph, and performing feature fusion on the standard user features and the standard product features in the standard feature mapping pair to obtain the standard recommendation features are:

[0031] Based on the user's historical behavior data, browsing history, and characteristic attributes of financial products, a financial product recommendation graph is constructed and candidate user feature groups are extracted from the financial product recommendation graph;

[0032] The candidate user feature groups and the candidate product feature groups are standardized, and the user feature groups and the product feature groups after the standardized processing are feature mapped using the preset mapping rules, and the expression is: Among them, U i is the i-th user feature in the candidate user feature group, P j is the jth product feature in the candidate product feature group, μ U is the user feature U i The mean value, σ U is the user feature U i The standard deviation of P Product feature P j The mean value, σ PProduct feature P j The standard deviation of and are the standardized user characteristics and product characteristics, M ij is the feature mapping score, which represents the standardized user features Product Features The degree of match between

[0033] The user features and product features in the generated standard feature mapping pairs are fused and the obtained standard recommendation features are optimized. The weights and importance of the recommended features are adjusted in combination with more contextual information. Finally, all the standard recommendation features are combined into a standard recommendation feature group as the basis for subsequent product recommendation value calculations.

[0034] Optionally, the steps of generating a standard recommended feature group according to all standard recommended features, and calculating a product recommendation value of a financial product according to the standard recommended feature group are:

[0035] All generated standard recommendation features are grouped together to form a complete standard recommendation feature group;

[0036] Assign weights to each feature in the standard recommended feature set based on the importance of the feature, the relevance of the feature to user behavior, and the analysis results of historical data;

[0037] Aggregate all features in the standard recommendation feature group, calculate the comprehensive score of each financial product, and calculate the product recommendation value for each financial product based on the comprehensive score;

[0038] All financial products are sorted according to the calculated recommendation values. Products with higher recommendation values ​​will be ranked at the top of the list. Financial products will be sorted according to the recommendation values ​​to generate a final recommendation list.

[0039] Optionally, the steps of recommending a financial product whose product recommendation value is greater than a recommendation threshold to a target financial user are:

[0040] Preset a recommendation threshold and filter the recommendation values ​​of all previously calculated financial products to select financial products with recommendation values ​​greater than the preset recommendation threshold;

[0041] The financial products that have passed the screening are grouped into a recommendation list and the screened recommended product list is further sorted;

[0042] Based on the screened list of recommended products, personalized recommendation content is generated for each target financial user, and recommended financial products are pushed to the target financial users through multiple channels based on the user's preferences and habits.

[0043] Optionally, after the user receives the recommendation information, the system monitors and records the user's response and behavior in real time and analyzes the user's selection rate, conversion rate, user feedback indicators and adjusts the recommendation strategy in real time according to the analysis results.

[0044] An intelligent financial product recommendation method comprises the following steps:

[0045] Generate a first financial product recommendation list based on facial image features of the person to be recommended;

[0046] Generate a second financial product recommendation list according to the time the recommended object browses the financial product details page;

[0047] Merging the first financial product recommendation list and the second financial product recommendation list to obtain a financial product recommendation list based on user implicit data;

[0048] Obtain a standard feature mapping pair based on the candidate user feature group and the candidate product feature group generated by the financial product recommendation graph, and perform feature fusion on the standard user features and the standard product features in the standard feature mapping pair to obtain a standard recommendation feature;

[0049] Generating a standard recommended feature group according to all standard recommended features, and calculating a product recommendation value of the financial product according to the standard recommended feature group;

[0050] Recommend financial products whose product recommendation values ​​are greater than the recommendation threshold to target financial users.

[0051] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below at the same time:

[0052] Through the user feature acquisition module and the user behavior analysis module, this application can comprehensively collect the user's explicit and implicit data, and generate two independent financial product recommendation lists. The list fusion module effectively integrates this data to generate a comprehensive recommendation list based on the user's multi-dimensional data. The standard feature mapping module further accurately matches user features with product features to ensure that the recommendation results are highly personalized and accurate to meet the personalized needs of users. In addition, there is a user behavior analysis module and a product recommendation value calculation module based on real-time data. Through real-time monitoring and analysis of user browsing behavior, the system can dynamically adjust the recommendation strategy and generate an optimized financial product recommendation list. This real-time response mechanism ensures that the recommendation results are highly consistent with the user's current needs and market conditions, improving user experience and satisfaction.

[0053] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The drawings described below are only some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0055] In the figure:

[0056] Figure 1 This is a block diagram of the intelligent financial product recommendation system.

[0057] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0058] The present invention will now be described in further detail with reference to the accompanying drawings.

[0059] See also Figure 1 As shown, in this embodiment, an intelligent financial product recommendation system is provided, including a user feature acquisition module, which is used to generate a first financial product recommendation list according to the facial image features of the object to be recommended;

[0060] The user behavior analysis module is used to generate a second list of financial product recommendations based on the time the recommended object browses the financial product details page; by integrating facial image features and browsing time, it can comprehensively utilize the user's explicit and implicit data to generate more accurate and personalized recommendation results. The integrated recommendation list can more comprehensively reflect the user's interests and needs, thereby improving the accuracy of the recommendation.

[0061] The list fusion module is used to merge the first financial product recommendation list and the second financial product recommendation list to obtain a financial product recommendation list based on the user's implicit data; dynamic adaptability enables the system to make flexible adjustments in different environments to maintain the accuracy and timeliness of the recommendation results. For example, when the market fluctuates, the system can automatically adjust the recommendation strategy to give priority to recommending low-risk products.

[0062] The standard feature mapping module is used to obtain standard feature mapping pairs based on the candidate user feature groups and candidate product feature groups generated by the financial product recommendation graph, and to fuse the standard user features and standard product features in the standard feature mapping pairs to obtain standard recommendation features; the use of mapping rules can effectively match user features with product features to generate standard recommendation features with a high degree of matching. The setting of the standard feature mapping module not only improves the computational efficiency, but also enhances the rationality and consistency of the recommendation effect.

[0063] The product recommendation value calculation module is used to generate a standard recommendation feature group based on all standard recommendation features, and calculate the product recommendation value of the financial product based on the standard recommendation feature group; by making full use of the user's behavior analysis results and standard feature mapping, it does not only rely on explicit data. This capability enables the system to dig deeper into user needs and provide more targeted recommendation services.

[0064] The financial product recommendation module is used to recommend financial products whose product recommendation values ​​are greater than a recommendation threshold to target financial users.

[0065] In this embodiment, the steps of generating the first financial product recommendation list according to the facial image features of the object to be recommended are:

[0066] Capturing a face image in real time through a camera, preprocessing the acquired face image, analyzing the preprocessed face image, and extracting a feature vector containing face features;

[0067] For example: In a bank's intelligent financial advisor application, when a user logs into the system through a camera, the system will capture the user's facial image in real time, perform feature analysis and matching, automatically identify the user's identity, and recommend the most suitable financial products based on the user's historical investment behavior and risk preferences.

[0068] Matching the extracted feature vector with the existing user feature database to identify the identity of the object to be recommended, and assigning the obtained identity of the object to be recommended to a specific user category;

[0069] For example, a user walks into a bank's automated financial advisor counter, and the camera automatically identifies the user and extracts the user's historical transaction data and risk preferences. The system immediately generates a personalized list of financial product recommendations and displays it on the screen, allowing the user to get customized financial advice without having to manually enter any information.

[0070] According to the facial image features and the feature matching results, a user profile of the recommended object is constructed, and based on the information in the user profile, financial products that best match the user profile are screened from the financial product library to generate a first financial product recommendation list;

[0071] For example, when a user logs into a financial service platform, the system uses a camera to identify the user and accurately recommends stock funds with high return potential based on their investment preferences, risk tolerance, and market trends, rather than low-risk conservative products. This personalized recommendation makes the user feel that the recommendation result is in line with their expectations.

[0072] The initially generated list of financial product recommendations is optimized and sorted, and finally a sorted list of first-tier financial product recommendations is formed based on the priority in the user portrait, the return-risk matching of the financial product, and the current market conditions. Through automated image processing, feature matching, and recommendation generation processes, the system greatly improves the fluidity of the user experience and the efficiency of the recommendation process. Users do not need to manually enter a large amount of information, and the system can quickly identify users and provide instant recommendation results, reducing users' waiting time and operational burden.

[0073] For example, in a banking application, users only need to stand in front of an ATM, and the system will automatically identify the user and recommend suitable financial products. Users no longer need to enter information and select options step by step. The whole process only takes a few seconds, which greatly saves time and improves the experience.

[0074] In this embodiment, the steps of generating the second financial product recommendation list according to the time duration of the recommended object browsing the financial product details page are:

[0075] Monitor and record the browsing behavior of the recommended objects on the financial product details page in real time and calculate the total stay time of the user on each financial product details page based on the collected behavior data. The expression is: Among them, T p is the total length of time a user stays on a financial product details page. n is the number of effective stays a user has made on the page. is the starting timestamp of the i-th entry into the page, that is, the time when the user enters the page. is the end timestamp of the i-th exit page, that is, the time when the user leaves the page, δ i is the effectiveness coefficient of the ith behavior. By real-time monitoring and recording the browsing behavior of users on the financial product details page and calculating the total length of time users stay on each page, the system can accurately capture the user's interests. The accuracy of this behavior data provides a reliable data basis for the subsequent interest rating and the construction of the recommendation list, ensuring that the recommendation results are more targeted.

[0076] For example, if a user stays on a high-risk investment product page for a long time and interacts with the page frequently, such as clicking to view detailed information and scrolling the page, the system will mark this product as a potential point of interest for the user, and give it priority in recommendations.

[0077] Statistical analysis is performed on the browsing time of users on different product detail pages to identify the financial products that users focus on and assign an interest score I to each browsed financial product based on the user's browsing time and behavior analysis results. p , whose expression is: Among them, I p: Interest score of financial product p. This value reflects the user's interest in the product relative to other products the user has browsed. b is the behavior weight, which indicates the influence of browsing time in calculating the interest score. is the sum of the weighted total stay time of the user on all the financial product detail pages that the user has browsed, where m represents the total number of financial products that the user has browsed, and T j represents the total length of time a user stays on the jth product page, w j represents the behavior weight of the jth product. p is the number of interactions when the user browses financial product p. Interaction behaviors may include clicking, scrolling, adding to favorites, consulting customer service, etc. It indicates the user's active attention to the product. max(C j ) is the maximum number of interactions that occurred on all viewed financial product pages. This value is used to normalize the number of interactions for comparison between different products. By combining browsing time and user interaction behaviors (such as clicks, scrolling, favorites, etc.), the system is able to assign a comprehensive interest score to each financial product. This score not only reflects the user's passive attention to the product (such as the length of stay), but also takes into account the user's active behavior (such as the number of interactions), thereby providing a multi-dimensional user interest assessment.

[0078] For example, for a user's behavior on different pages, the system not only calculates the length of time they stay, but also takes into account the user's interactive behavior. If a user clicks on a fund product page multiple times and views related information, the interest score of the product will increase significantly.

[0079] Combined with the user's interest score, a user's product preference model is constructed and based on the user's product preference model, a group of candidate products that best match the user's preferences are screened from the financial product library, which is a preliminary second financial product recommendation list; the product preference model integrates the user's interest in different products, thereby being able to more accurately reflect the user's needs and preferences.

[0080] The selected candidate products are optimized and sorted, and finally an optimized second financial product recommendation list is generated. By comprehensively analyzing the user's behavioral data, interest ratings, and product preference models, the system is able to generate an accurate and optimized list of financial product recommendations. This accuracy not only improves the relevance of recommendations, but also significantly enhances user satisfaction, because users can receive recommended products that better meet their needs. Secondly, through detailed monitoring and analysis of user behavior, the system demonstrates a high level of intelligence. Users do not need to manually enter or select preferences. The system can automatically identify and understand user needs and generate corresponding recommendations. This greatly improves the user experience and ease of use of the system.

[0081] In this embodiment, the steps of merging the first financial product recommendation list and the second financial product recommendation list to obtain a financial product recommendation list based on user implicit data are:

[0082] Receiving financial product data in a first financial product recommendation list and a second financial product recommendation list;

[0083] The products in the first financial product recommendation list and the second financial product recommendation list are weighted according to the system preset rules; the system preset rules are a set of strategies and parameters set according to user needs, market conditions and business logic to guide the weight allocation and priority score calculation. Then, the weight allocation results of the same financial product from the two lists are weighted and summed to calculate the comprehensive priority score of each product. For the same product appearing in both lists, its comprehensive score will be improved, and for the product appearing in only one list, the comprehensive score will be determined based on the weight of the list in which it is located;

[0084] When the same financial products exist in the two lists, the ranking of the products in the merged list is determined according to the priority score;

[0085] The financial products that have been through priority score calculation and conflict resolution are merged to form a unified financial product recommendation list. The merged list will include the financial products in the two recommendation lists and will be sorted by priority score;

[0086] Based on the user's personalized preferences and historical behavior data, the integrated financial product recommendation list is further optimized, and finally a financial product recommendation list based on the user's implicit data is generated.

[0087] In this embodiment, the steps of obtaining a standard feature mapping pair based on the candidate user feature group and the candidate product feature group generated by the financial product recommendation graph, and fusing the standard user features and the standard product features in the standard feature mapping pair to obtain the standard recommendation features are as follows:

[0088] Based on the user's historical behavior data, browsing history, and characteristic attributes of financial products, a financial product recommendation graph is constructed and candidate user feature groups are extracted from the financial product recommendation graph;

[0089] The candidate user feature groups and the candidate product feature groups are standardized, and the user feature groups and the product feature groups after the standardized processing are feature mapped using the preset mapping rules, and the expression is: Among them, U i : The i-th user feature in the candidate user feature group, P j : The jth product feature in the candidate product feature group. μ U :User characteristics Ui The mean of is calculated based on historical data or training data set. U is the user feature U i The standard deviation of P Product feature P j The mean value, σ P Product feature P j The standard deviation of and It is the user features and product features after standardization, indicating the deviation of the feature value relative to the mean, normalized by the standard deviation. ij is the feature mapping score, which represents the standardized user features Product Features The mapping rules include risk tolerance and product risk level mapping rules, user age and product suitability mapping rules, investment preference and product type mapping rules, user financial status and investment scale mapping rules, user historical behavior and product attributes mapping rules, time sensitivity and product timeliness mapping rules, and market dynamics and product recommendation mapping rules. The rule for mapping risk tolerance and product risk level is to match the user's risk tolerance with the risk level of the financial product. For example, users with high risk tolerance should match high-risk and high-yield products, while users with low risk tolerance should match low-risk and stable-yield products. The rule for mapping investment preference and product type is to match the user's investment preference (such as long-term investment, short-term speculation, environmental protection investment, etc.) with the type of financial product. For example, users who prefer long-term stable returns should recommend products with long-term investment value, while users who prefer short-term speculation should recommend short-term products with high liquidity. The rule for mapping user age and product suitability is to recommend financial products suitable for the age group according to the user's age stage. For example, young users may prefer high-risk and high-yield investment products, while older users may pay more attention to security and value-preserving products. The mapping rule between user financial status and investment scale is to match the user's financial status (such as disposable income, asset size) with the minimum investment amount and investment scale of the investment product. Users with good financial status can recommend high-threshold high-quality investment products, while users with average financial status should recommend products with lower thresholds. The mapping rule between user historical behavior and product attributes is to match the user's historical behavior (such as financial products purchased in the past, browsing habits) with the attributes of financial products. For example, the types of products that users have purchased in the past may reflect their preferences, and products with similar attributes are recommended.

[0090] The user features and product features in the generated standard feature mapping pairs are fused and the obtained standard recommendation features are optimized. The weights and importance of the recommendation features are adjusted in combination with more contextual information. Finally, all the standard recommendation features are combined into a standard recommendation feature group as the basis for the subsequent product recommendation value calculation. By building a recommendation graph based on the user's historical behavior data and financial product features, the system can accurately capture the user's preferences and needs. Standardization and feature mapping ensure the precise match between user features and product features. Combined with multiple mapping rules (such as risk tolerance, investment preferences, user age, etc.), the system can provide personalized recommendations based on the specific circumstances of different users. Feature fusion and optimization steps further enhance the accuracy and flexibility of recommendations. By dynamically adjusting weights, the system can adapt to changes in users and the market and ensure the relevance and timeliness of recommendation results. Finally, all standard recommendation features are combined into a complete feature group, which lays a solid foundation for the subsequent recommendation value calculation, ensuring that the system can generate accurate and personalized financial product recommendations and improve user satisfaction and conversion rate.

[0091] In this embodiment, the steps of generating a standard recommendation feature group according to all standard recommendation features, and calculating the product recommendation value of the financial product according to the standard recommendation feature group are as follows:

[0092] All generated standard recommendation features are grouped together to form a complete standard recommendation feature group;

[0093] Assign weights to each feature in the standard recommended feature set based on the importance of the feature, the relevance of the feature to user behavior, and the analysis results of historical data;

[0094] Aggregate all features in the standard recommendation feature group, calculate the comprehensive score of each financial product, and calculate the product recommendation value for each financial product based on the comprehensive score;

[0095] All financial products are sorted according to the calculated recommendation value. Products with higher recommendation values ​​will be ranked at the top of the list. The financial products will be sorted according to the recommendation value to generate the final recommendation list. By assigning weights to features, the system can accurately reflect the real needs of users based on the importance and relevance of features. Aggregating features and calculating comprehensive scores ensures that the recommendation value of each financial product is based on a comprehensive analysis, making the final recommendation results more objective and accurate. By sorting to generate the final recommendation list, the system can prioritize the products that are most suitable for users, improving the effectiveness of recommendations and user satisfaction.

[0096] In this embodiment, the steps of recommending a financial product whose product recommendation value is greater than the recommendation threshold to a target financial user are:

[0097] A recommendation threshold is preset and the recommendation values ​​of all previously calculated financial products are screened to select financial products whose recommendation values ​​are greater than the preset recommendation threshold. By screening the recommendation values ​​of all financial products, the system can filter out products that do not meet user needs and retain only high-quality options.

[0098] The financial products that have passed the screening are grouped into a recommendation list and the screened recommended product list is further sorted;

[0099] Based on the screened recommended product list, personalized recommendation content is generated for each target financial user and recommended financial products are pushed to the target financial user through multiple channels based on the user's preferences and habits. By generating personalized recommendation content and using multi-channel push, the system can provide more targeted and efficient recommendation services based on user preferences and habits, improving user experience and conversion rate. This comprehensive recommendation strategy not only optimizes the quality of recommendations, but also enhances the relevance of recommendations and user satisfaction.

[0100] In this embodiment, after the user receives the recommendation information, the system monitors and records the user's reaction and behavior in real time, analyzes the user's selection rate, conversion rate, and user feedback indicators, and adjusts the recommendation strategy in real time according to the analysis results. By monitoring the user's reaction and behavior after receiving the recommendation information in real time, the system can accurately capture the user's selection rate, conversion rate and other key feedback indicators. Based on these data analyses, the system can quickly adjust and optimize the recommendation strategy, thereby improving the accuracy of the recommendation and user satisfaction. This closed-loop feedback mechanism not only improves the intelligence level of the system, but also ensures that the recommendation service can be continuously optimized, dynamically respond to user needs and market changes, and provide users with a more personalized and efficient recommendation experience.

[0101] An intelligent financial product recommendation method comprises the following steps:

[0102] Generate a first financial product recommendation list based on facial image features of the person to be recommended;

[0103] Generate a second financial product recommendation list according to the time the recommended object browses the financial product details page;

[0104] Merging the first financial product recommendation list and the second financial product recommendation list to obtain a financial product recommendation list based on user implicit data;

[0105] Obtain a standard feature mapping pair based on the candidate user feature group and the candidate product feature group generated by the financial product recommendation graph, and perform feature fusion on the standard user features and the standard product features in the standard feature mapping pair to obtain a standard recommendation feature;

[0106] Generating a standard recommended feature group according to all standard recommended features, and calculating a product recommendation value of the financial product according to the standard recommended feature group;

[0107] Recommend financial products whose product recommendation values ​​are greater than the recommendation threshold to target financial users.

[0108] The present invention is not limited to the above-mentioned embodiments. Anyone should be aware that any structural changes made under the enlightenment of the present invention, and any technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention. The technology, shape, and structural parts not described in detail in the present invention are all well-known technologies.

Claims

1. An intelligent financial product recommendation system, characterized in that: include: A user feature acquisition module, the user feature acquisition module is used to generate a first financial product recommendation list according to the facial image features of the object to be recommended; A user behavior analysis module, the user behavior analysis module is used to generate a second financial product recommendation list according to the length of time that the recommended object browses the financial product details page; A list fusion module, which is used to fuse the first financial product recommendation list and the second financial product recommendation list to obtain a financial product recommendation list based on user implicit data; A standard feature mapping module, which is used to obtain a standard feature mapping pair according to a candidate user feature group and a candidate product feature group generated by a financial product recommendation graph, and to fuse the standard user features and the standard product features in the standard feature mapping pair to obtain a standard recommendation feature; A product recommendation value calculation module, which is used to generate a standard recommendation feature group according to all standard recommendation features, and calculate the product recommendation value of the financial product according to the standard recommendation feature group; The financial product recommendation module is used to recommend financial products whose product recommendation values ​​are greater than the recommendation threshold to target financial users.

2. The intelligent financial product recommendation system according to claim 1, characterized in that: The steps of generating the first financial product recommendation list according to the facial image features of the recommended object are: Capturing a face image in real time through a camera, preprocessing the acquired face image, analyzing the preprocessed face image, and extracting a feature vector containing face features; Matching the extracted feature vector with the existing user feature database to identify the identity of the object to be recommended, and assigning the obtained identity of the object to be recommended to a specific user category; According to the facial image features and the feature matching results, a user profile of the recommended object is constructed, and based on the information in the user profile, financial products that best match the user profile are screened from the financial product library to generate a first financial product recommendation list; The initially generated financial product recommendation list is optimized and sorted, and finally a sorted first financial product recommendation list is formed based on the priority in the user portrait, the return-risk matching of the financial products and the current market conditions.

3. The intelligent financial product recommendation system according to claim 1, characterized in that: The steps of generating the second financial product recommendation list according to the time that the recommended object browses the financial product details page are as follows: Monitor and record the browsing behavior of the recommended objects on the financial product details page in real time and calculate the total stay time of the user on each financial product details page based on the collected behavior data. The expression is: Among them, T p is the total length of time a user stays on a financial product details page, n is the number of effective stays performed by the user on the page, is the starting timestamp of the i-th entry into the page, that is, the time when the user enters the page, is the end timestamp of the i-th exit page, that is, the time when the user leaves the page, δ i is the effectiveness coefficient of the ith behavior; Perform statistical analysis on the browsing time of users on different product detail pages, identify the financial products that users focus on, and assign an interest score U to each browsed financial product based on the user's browsing time and behavior analysis results. p , whose expression is: Among them, I p : Interest score of financial product p, w b is the behavior weight, is the sum of the weighted total stay time of the user on all the financial product detail pages that the user has browsed, m represents the total number of financial products that the user has browsed, and T j represents the total length of time a user stays on the jth product page, w j represents the behavior weight of the jth product, C p is the number of interactions of users when browsing financial product p, max(C j ) is the maximum number of interactions that occurred on all financial product pages viewed by the user; Combined with the user's interest score, a user's product preference model is constructed, and based on the user's product preference model, a group of candidate products that best match the user's preferences are screened from the financial product library to form a preliminary second financial product recommendation list; The screened candidate products are optimized and sorted, and finally an optimized second financial product recommendation list is generated.

4. The intelligent financial product recommendation system according to claim 1, characterized in that: The steps of merging the first financial product recommendation list and the second financial product recommendation list to obtain a financial product recommendation list based on user implicit data are: Receiving financial product data in a first financial product recommendation list and a second financial product recommendation list; Assign weights to the products in the first financial product recommendation list and the second financial product recommendation list according to the system's preset rules. Then, perform a weighted summation of the weight assignment results of the same financial product from the two lists to calculate the comprehensive priority score of each product. For the same product that appears in both lists, its comprehensive score will be increased, while for the product that appears in only one list, the comprehensive score will be determined based on the weight of the list to which it belongs; When the same financial products exist in the two lists, the ranking of the products in the merged list is determined according to the priority score; The financial products that have been through priority score calculation and conflict resolution are merged to form a unified financial product recommendation list. The merged list will include the financial products in the two recommendation lists and will be sorted by priority score; Based on the user's personalized preferences and historical behavior data, the integrated financial product recommendation list is further optimized, and finally a financial product recommendation list based on the user's implicit data is generated.

5. The intelligent financial product recommendation system according to claim 1, characterized in that: The steps of obtaining a standard feature mapping pair based on the candidate user feature group and the candidate product feature group generated by the financial product recommendation graph, and fusing the standard user features and the standard product features in the standard feature mapping pair to obtain the standard recommendation features are as follows: Based on the user's historical behavior data, browsing history, and characteristic attributes of financial products, a financial product recommendation graph is constructed and candidate user feature groups are extracted from the financial product recommendation graph; The candidate user feature groups and the candidate product feature groups are standardized, and the user feature groups and the product feature groups after the standardized processing are feature mapped using the preset mapping rules, and the expression is: Among them, U i is the i-th user feature in the candidate user feature group, P j is the jth product feature in the candidate product feature group, μ U is the user feature U i The mean value, σ U is the user feature U i The standard deviation of P Product feature P j The mean value, σ P Product feature P j The standard deviation of and are the standardized user characteristics and product characteristics, M ij is the feature mapping score, which represents the standardized user features Product Features The degree of match between The user features and product features in the generated standard feature mapping pair are fused and the obtained standard recommendation features are optimized. The weight and importance of the recommended features are adjusted in combination with more contextual information. Finally, all the standard recommendation features are combined into a standard recommendation feature group.

6. The intelligent financial product recommendation system according to claim 1, characterized in that: The steps of generating a standard recommended feature group according to all standard recommended features and calculating the product recommendation value of the financial product according to the standard recommended feature group are as follows: All generated standard recommendation features are grouped together to form a complete standard recommendation feature group; Assign weights to each feature in the standard recommended feature set based on the importance of the feature, the relevance of the feature to user behavior, and the analysis results of historical data; Aggregate all features in the standard recommendation feature group, calculate the comprehensive score of each financial product, and calculate the product recommendation value for each financial product based on the comprehensive score; All financial products are sorted according to the calculated recommendation values. Products with higher recommendation values ​​will be ranked at the top of the list. Financial products will be sorted according to the recommendation values ​​to generate a final recommendation list.

7. The intelligent financial product recommendation system according to claim 1, characterized in that: The steps of recommending financial products whose product recommendation value is greater than the recommendation threshold to target financial users are: Preset a recommendation threshold and filter the recommendation values ​​of all previously calculated financial products to select financial products with recommendation values ​​greater than the preset recommendation threshold; The financial products that have passed the screening are grouped into a recommendation list and the screened recommended product list is further sorted; Based on the screened list of recommended products, personalized recommendation content is generated for each target financial user, and recommended financial products are pushed to the target financial users through multiple channels based on the user's preferences and habits.

8. The intelligent financial product recommendation system according to claim 1, characterized in that: After the user receives the recommendation information, the system monitors and records the user's response and behavior in real time, analyzes the user's selection rate, conversion rate, and user feedback indicators, and adjusts the recommendation strategy in real time based on the analysis results.

9. An intelligent financial product recommendation method, characterized in that: The following steps are involved: Generate a first financial product recommendation list based on facial image features of the person to be recommended; Generate a second financial product recommendation list according to the time the recommended object browses the financial product details page; Merging the first financial product recommendation list and the second financial product recommendation list to obtain a financial product recommendation list based on user implicit data; Obtain a standard feature mapping pair based on the candidate user feature group and the candidate product feature group generated by the financial product recommendation graph, and perform feature fusion on the standard user features and the standard product features in the standard feature mapping pair to obtain a standard recommendation feature; Generating a standard recommended feature group according to all standard recommended features, and calculating a product recommendation value of the financial product according to the standard recommended feature group; Recommend financial products whose product recommendation values ​​are greater than the recommendation threshold to target financial users.

Images