A personalized recommendation method and system for corporate financial products based on preference analysis

Through a personalized recommendation method based on preference analysis, combined with GBDT model and preference analysis, a company portrait and different preference attribute tree is constructed, which solves the problem that enterprises find suitable for financial products, and achieves accurate recommendation and resource optimization.

CN118657592BActive Publication Date: 2024-11-29JINAN HIGH-TECH INNOVATION INVESTMENT GRP CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411147085.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-21
Publication Date
2024-11-29
Estimated Expiration
2044-08-21

AI Technical Summary

Technical Problem

It is difficult for enterprises to find the most suitable product when choosing financial products. The existing recommendation methods are too concentrated to effectively match corporate needs, affecting user experience and resource utilization.

Method used

Using a personalized recommendation method based on preference analysis, the company inputs information to generate preference description text, combines the GBDT model and preference analysis model to build a corporate portrait, matches the historical recommendation matrix and financial product characteristics, generates a different preference attribute tree, and groups the preference attributes to Match financial products.

Benefits of technology

It has achieved accurate financial product recommendations, improved user experience, enhanced user stickiness, helped enterprises quickly find products that suit their needs, and optimized the resource allocation of financial institutions to understand market demand.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118657592B_ABST
    Figure CN118657592B_ABST
Patent Text Reader

Abstract

The present application provides a method and system for personalized recommendation of corporate financial products based on preference analysis, which belongs to the field of data processing technology. The method determines the preference attribute set of the enterprise to be recommended based on the enterprise preference description text group and the preference analysis model. Determine the financial business data of the enterprise to be recommended and generate the enterprise portrait of the enterprise to be recommended based on the financial business data and the GBDT model. Based on the preset portrait matching rules, match the historical product recommendation matrix of the historical recommended enterprise set corresponding to the enterprise portrait. Based on the historical product recommendation matrix and the preset financial product feature matrix, generate a recommended difference product set and a recommended difference preference attribute tree. Based on the preference attribute set, the recommended difference preference attribute tree and the preference matching degree recognition model, determine the preference attribute grouping of the enterprise to be recommended, and match each financial product to be recommended in the financial product library based on the preference attribute grouping and the enterprise portrait, so as to obtain a recommended financial product sequence and send it to the enterprise terminal.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and system for personalized recommendation of corporate financial products based on preference analysis. Background Art

[0002] At present, the number of corporate financial products developed is increasing, and the number and types of enterprises are constantly increasing. When choosing financial products suitable for their own enterprises, enterprise personnel often rely on subjective judgment or use evaluations collected through other channels to make choices. Faced with a large number of financial products, it is very difficult to choose the most suitable financial product. Generally, when recommending financial products to enterprises, relevant financial products will be recommended based on the enterprise's purchase records or the applicable enterprises pre-set for the financial products. This makes it impossible for enterprises to obtain the most suitable financial products for themselves. At the same time, it may cause the recommended financial products to be too concentrated, and some financial products cannot be effectively matched with enterprises and recommended. Summary of the invention

[0003] The embodiments of the present application provide a method and system for personalized recommendation of corporate financial products based on preference analysis, which is used to solve the problem that financial products cannot be effectively and accurately recommended to corporate users, affecting the system usage experience of corporate users, and financial product resources cannot be reasonably utilized.

[0004] On the one hand, an embodiment of the present application provides a method for personalized recommendation of corporate financial products based on preference analysis, the method comprising:

[0005] Determine a corresponding enterprise preference description text group based on enterprise input information from an enterprise terminal;

[0006] Determine the preference attribute set of the enterprise to be recommended according to the enterprise preference description text group and the preset preference analysis model;

[0007] Determine the financial business data of the enterprise to be recommended, and generate a corporate profile of the enterprise to be recommended based on the financial business data and the pre-trained GBDT model; wherein the financial business data at least includes the capital flow information, credit record information, historical interaction behavior with financial products, and preset corporate scientific and technological innovation capabilities of the enterprise to be recommended;

[0008] Based on the preset portrait matching rules, the historical product recommendation matrix of the historical recommended enterprise set corresponding to the enterprise portrait is matched; wherein the preset portrait matching rules are used to compare the enterprise portrait with the recommended enterprise portrait corresponding to each historically recommended enterprise;

[0009] Generate a corresponding recommended difference product set and its corresponding recommended difference preference attribute tree according to the historical product recommendation matrix and the preset financial product feature matrix;

[0010] Based on the preference attribute set, the recommendation difference preference attribute tree and the preset preference matching degree identification model, the preference attribute grouping of the enterprise to be recommended is determined, so as to match each financial product to be recommended in the financial product library based on the preference attribute grouping and the enterprise portrait, so as to obtain a recommended financial product sequence and send it to the enterprise terminal.

[0011] In one implementation of the present application, based on the enterprise input information from the enterprise terminal, determining the corresponding enterprise preference description text group specifically includes:

[0012] Inputting the enterprise input information from the enterprise terminal into a natural language processing model to perform word segmentation processing on the enterprise input information to obtain a plurality of pending preference description texts;

[0013] Encoding each of the pending preference description texts to generate a corresponding pending preference description text vector;

[0014] Calculating the product of the pending preference description text vector with the first vector and the second vector respectively, so as to compare the corresponding first product value and the second product value with the preset completeness evaluation standard respectively; wherein the first vector is obtained based on a number of basic preference description keywords; and the second vector is obtained based on a number of specific preference description keywords;

[0015] According to the comparison results, determine the completeness of the preference description corresponding to the enterprise input information;

[0016] When the completeness of the preference description is greater than a predetermined value, each of the pending preference description texts is added to the enterprise preference description text group;

[0017] Otherwise, based on the comparison result, corresponding supplementary prompt information for entry of basic preference dimensions and / or specific preference dimensions is generated, and the supplementary prompt information for entry is sent to the enterprise terminal.

[0018] In one implementation of the present application, the preference attribute set of the enterprise to be recommended is determined according to the enterprise preference description text group and the preset preference analysis model, specifically including:

[0019] Determine a preference text vector corresponding to the enterprise preference description text group;

[0020] Inputting the preference text vector into the preference analysis model to determine the corresponding preference attributes and their corresponding prediction scores; wherein the preference analysis model is a pre-trained logistic regression model for identifying preference attributes;

[0021] Each of the preference attributes and the corresponding prediction score are added to the preference attribute set.

[0022] In one implementation of the present application, generating the enterprise portrait of the enterprise to be recommended according to the financial business data and the pre-trained GBDT model specifically includes:

[0023] Generate a business data matrix corresponding to the financial business data according to the data parameter type; the rows of the business data matrix correspond to different data parameter types, and the columns correspond to different times;

[0024] Input the business data matrix into the GBDT model to perform iterative operations until a preset number of iterations is reached, and obtain the portrait indicators of multiple enterprise labels corresponding to the financial business data; the enterprise labels include at least one or more of the following: enterprise type, development stage, and risk level;

[0025] According to the portrait indicators, a corporate portrait corresponding to the to-be-recommended enterprise is constructed.

[0026] In one implementation of the present application, based on a preset portrait matching rule, a historical product recommendation matrix of a historically recommended enterprise set corresponding to the enterprise portrait is matched, specifically including:

[0027] Calculating the similarity between the enterprise portrait and the recommended enterprise portraits of each of the historically recommended enterprises in a preset database; wherein the similarity is calculated based on the feature vectors corresponding to the enterprise portrait and the recommended enterprise portrait respectively;

[0028] Adding each of the historically recommended enterprises whose portrait similarity is greater than a preset similarity threshold to the historically recommended enterprise set;

[0029] The historical product recommendation matrix corresponding to the enterprise portrait is calculated according to the historical recommended enterprise set, the portrait similarity and the preset initial historical product recommendation matrix; the rows of the historical product recommendation matrix are the historically recommended enterprises, which are listed as recommended historical products.

[0030] In one implementation of the present application, a corresponding recommended difference product set and its corresponding recommended difference preference attribute tree are generated according to the historical product recommendation matrix and the preset financial product feature matrix, specifically including:

[0031] Calculate the recommended feature values ​​of each of the historical products recommended by each of the historically recommended enterprises according to the historical product recommendation matrix and the preset financial product feature matrix;

[0032] Determining a recommended average feature vector corresponding to each of the historical products according to the recommended feature value corresponding to each of the historical products and the historical product recommendation matrix;

[0033] Determine a corresponding recommendation difference matrix according to the vector differences between the preset financial product feature matrix and each of the recommended average feature vectors;

[0034] Compare each element in the recommendation difference matrix with a preset recommendation threshold value, so as to filter the historical products that meet the preset conditions according to the comparison results, and add them to the recommended difference product set;

[0035] Each product feature corresponding to the recommended difference product set is used as an attribute tree node to construct the corresponding recommended difference preference attribute tree.

[0036] In one implementation of the present application, based on the preference attribute set, the recommendation difference preference attribute tree and the preset preference matching degree recognition model, determining the preference attribute grouping of the enterprise to be recommended specifically includes:

[0037] Through the preset preference matching degree recognition model, each preference attribute in the preference attribute set is traversed through the recommendation difference preference attribute tree one by one to determine a matching attribute tree node of each preference attribute; the matching degree corresponding to the matching attribute tree node is greater than a preset preference matching degree threshold;

[0038] According to the matching attribute tree nodes, the preference attributes are divided into the preference attribute groups, and the preference attributes are removed from the preference attribute set until the preference attribute set is an empty set.

[0039] In one implementation of the present application, before matching each to-be-recommended financial product in the financial product library based on the preference attribute grouping and the enterprise portrait to obtain a recommended financial product sequence and sending it to the enterprise terminal, the method further includes:

[0040] Determine the corresponding similar evaluation information in a preset evaluation list according to the enterprise portrait; wherein the preset evaluation list includes scores of several enterprises on each of the financial products to be recommended; and the similar evaluation information includes scores of the enterprises corresponding to the enterprise portrait on each of the financial products to be recommended;

[0041] Calculating the evaluation similarity between the financial products to be recommended according to the similar evaluation information, the preset evaluation list and the preset collaborative filtering algorithm;

[0042] According to the similarities of each of the evaluations, a recommendation score corresponding to each of the financial products to be recommended is determined, so that each of the financial products to be recommended whose recommendation score is greater than a score threshold is regarded as the financial product to be recommended that matches the corporate profile.

[0043] In one implementation of the present application, matching each to-be-recommended financial product in the financial product library based on the preference attribute grouping and the enterprise portrait to obtain a recommended financial product sequence and send it to the enterprise terminal specifically includes:

[0044] According to the preference attribute grouping, the financial products to be recommended that match the corporate profile are screened to obtain recommended financial products;

[0045] According to the preset sequence length, each of the recommended financial products is sequentially added to the recommended financial product sequence of the preset sequence length in descending order of the recommendation score.

[0046] The present application also provides a personalized recommendation system for corporate financial products based on preference analysis, the system comprising:

[0047] A first determination module, used to determine a corresponding enterprise preference description text group based on enterprise input information from an enterprise terminal;

[0048] A second determination module is used to determine a set of preference attributes of the enterprise to be recommended according to the enterprise preference description text group and a preset preference analysis model;

[0049] A determination generation module is used to determine the financial business data of the enterprise to be recommended, and generate a corporate portrait of the enterprise to be recommended based on the financial business data and the pre-trained GBDT model; wherein the financial business data at least includes the capital flow information, credit record information, historical interaction behavior with financial products, and preset corporate scientific and technological innovation capabilities of the enterprise to be recommended;

[0050] A matching module, for matching the historical product recommendation matrix of the historically recommended enterprise set corresponding to the enterprise portrait based on a preset portrait matching rule; wherein the preset portrait matching rule is used to compare the enterprise portrait with the recommended enterprise portrait corresponding to each historically recommended enterprise;

[0051] A generation module, used to generate a corresponding set of recommended difference products and a corresponding recommended difference preference attribute tree according to the historical product recommendation matrix and the preset financial product feature matrix;

[0052] The third determination module is used to determine the preference attribute grouping of the enterprise to be recommended based on the preference attribute set, the recommendation difference preference attribute tree and the preset preference matching degree recognition model, so as to match each financial product to be recommended in the financial product library based on the preference attribute grouping and the enterprise portrait, so as to obtain a recommended financial product sequence and send it to the enterprise terminal.

[0053] Compared with the prior art, the present invention has the following significant effects:

[0054] 1) Through the above scheme, this application can deeply analyze the preference description text entered by the enterprise, combine the preference analysis model to obtain the preference attribute set corresponding to the personalized needs of the enterprise, and use the GBDT model to analyze the enterprise portrait, which helps to deeply understand the actual situation of the enterprise in order to accurately recommend financial products.

[0055] 2) By using the preset portrait matching rules, we can quickly find historical recommendation cases with high similarity, and further obtain the recommendation difference preference attribute tree that can provide targeted and differentiated recommendations. Based on the preference attribute set, the recommendation difference preference attribute tree and the preference matching recognition model, we can group the enterprise preference attributes to match the financial products to be recommended, realize personalized recommendations, significantly improve the enterprise's user experience, enhance user stickiness, and promote the effective promotion of financial products.

[0056] 3) This application can help enterprises quickly find financial products that suit their needs, guide financial institutions to optimize resource allocation, and help financial institutions better understand market demand and develop financial products that better meet market demand. This will overcome the current problems that financial products cannot be effectively and accurately recommended to corporate users, affecting the system experience of corporate users, and that financial product resources cannot be reasonably utilized. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0058] Figure 1 A flowchart of a method for personalized recommendation of corporate financial products based on preference analysis in an embodiment of the present application;

[0059] Figure 2 This is a schematic diagram of the structure of a personalized recommendation system for corporate financial products based on preference analysis in an embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0061] The embodiments of the present application provide a method and system for personalized recommendation of corporate financial products based on preference analysis, which is used to solve the problem that financial products cannot be effectively and accurately recommended to corporate users, affecting the system usage experience of corporate users, and the problem that financial product resources cannot be reasonably utilized. Among them, the financial products listed in this application all have product labels pre-marked by financial institutions, and preliminary product recommendations can be made for enterprises through product labels and corporate portraits. The users of this application include the enterprise side and the financial institution side. Financial institutions can list financial products in the corresponding system of this application, and the enterprise side uses the system to select the financial products that are suitable for them.

[0062] The following describes in detail various embodiments of the present application in conjunction with the accompanying drawings.

[0063] The present application embodiment provides a method for personalized recommendation of corporate financial products based on preference analysis, such as Figure 1 As shown, the method may include steps S101-S106:

[0064] S101, the server determines a corresponding enterprise preference description text group based on enterprise input information from an enterprise terminal.

[0065] It should be noted that the server, as the executor of the method for personalized recommendation of corporate financial products based on preference analysis, is only an example, and the executor is not limited to the server, and this application does not make specific restrictions on this. The corporate terminal can be a mobile phone, computer or other device of relevant personnel within the company to be recommended financial products, and this application does not make specific restrictions on this.

[0066] Enterprise users can send relevant information to the server through the enterprise terminal, such as registration account information, enterprise type, enterprise establishment time, whether it is a specialized, refined, special and new enterprise, and enterprise certification (high-tech enterprise), etc. Enterprise input information can be text used to express the current status and future plans of the enterprise, and to describe preferences for financial products.

[0067] In the embodiment of the present application, based on the enterprise input information from the enterprise terminal, the corresponding enterprise preference description text group is determined, specifically including:

[0068] First, the enterprise input information from the enterprise terminal is input into the natural language processing model to perform word segmentation processing on the enterprise input information to obtain multiple pending preference description texts. Secondly, each pending preference description text is encoded to generate a corresponding pending preference description text vector. Subsequently, the product between the pending preference description text vector and the first vector and the second vector is calculated to compare the corresponding first product value and the second product value with the preset completeness evaluation standard. Among them, the first vector is obtained based on several basic preference description keywords. The second vector is obtained based on several specific preference description keywords. Then, according to the comparison result, the completeness of the preference description corresponding to the enterprise input information is determined. When the completeness of the preference description is greater than a predetermined value, each pending preference description text is added to the enterprise preference description text group. Otherwise, based on the comparison result, the corresponding basic preference dimension and / or specific preference dimension input supplementary prompt information is generated, and the input supplementary prompt information is sent to the enterprise terminal.

[0069] That is to say, the present application combines the natural language processing model to efficiently process the enterprise input information and extract the pending preference description text related to the preference for financial products in the enterprise input information. Among them, the natural language processing model can be a trained rule-based word segmentation, statistical word segmentation and other algorithm models. Through the natural language processing model, the enterprise input information can be divided into a number of meaningful words and phrases to obtain the pending preference description text. After obtaining the pending preference description text, the server vectorizes the text to generate the pending preference description text vector for calculation and analysis.

[0070] The undetermined preference description text vector is multiplied with the preset vectors, namely the first vector and the second vector, respectively, to obtain the first product value and the second product value. Among them, the first vector and the second vector represent different preference standards. The first vector represents the preference standard for basic preferences. The basic preferences can be understood as the basic requirement preference points of all enterprises for financial product preferences; the second vector represents the preference standard for specific preferences. The specific preferences can be understood as the requirements of various enterprises in the market for individual attribute preferences of financial products. The first vector and the second vector can be set by the user based on historical data or expert experience, and this application does not make specific restrictions on this. The preset completeness evaluation standard includes a preset list of the completeness of the pre-set evaluation text preference description. For the first product value and the second product value, there is a list column in the preset completeness evaluation standard for comparison.

[0071] The first product value is compared with the preset list of preset completeness evaluation standards to obtain the first evaluation value in the list, and the second product value is compared with the preset list of preset completeness evaluation standards to obtain the second evaluation value. The server inputs the first evaluation value and the second evaluation value into the preference description completeness calculation function, and the calculation function can weight the first evaluation value and the second evaluation value to obtain the corresponding function value, which corresponds to the completeness of the preference description. If the completeness of the preference description is greater than the predetermined value set by the user, the server adds the pending preference description text to the enterprise preference description text group. Otherwise, it means that the completeness of the preference description is insufficient, and the enterprise input information needs to be supplemented before the enterprise's preference attributes can be clarified. Then the server determines whether the first evaluation value and the second evaluation value are greater than the supplementary threshold, and uses the evaluation value that is not greater than the supplementary threshold as the dimension to be supplemented, such as the basic preference dimension and the specific preference dimension. The server generates corresponding supplementary prompt information and sends it to the enterprise terminal so that the user of the enterprise terminal can supplement the information for the dimension to be supplemented.

[0072] This application combines natural language processing technology to conduct in-depth analysis of enterprise input information and judge the completeness of preference description by calculating vector products. It can improve the degree of automation of information processing and enhance the accuracy and objectivity of preference evaluation, so as to accurately recommend financial products based on enterprise preferences.

[0073] S102, the server determines a set of preference attributes of the enterprises to be recommended according to the enterprise preference description text group and a preset preference analysis model.

[0074] In the embodiment of the present application, the preference attribute set of the enterprise to be recommended is determined according to the enterprise preference description text group and the preset preference analysis model, specifically including:

[0075] The server determines the preference text vector corresponding to the enterprise preference description text group, inputs the preference text vector into the preference analysis model, and determines the corresponding preference attributes and their corresponding prediction scores. The preference analysis model is a pre-trained logistic regression model for identifying preference attributes. Each preference attribute and its corresponding prediction score are added to the preference attribute set.

[0076] That is, the server converts the enterprise preference description text group into a vector form, such as the enterprise preference description text group is ,in It is Business preference description text, , the preference text vector is The preference analysis model is a logistic regression model trained by a number of preference vector samples. The preference vector samples at least contain preference attributes and predicted score labels. The trained preference analysis model can identify each preference attribute corresponding to the preference text vector and its predicted score. Afterwards, the server adds the model output results, i.e., each preference attribute corresponding to the preference text vector and its predicted score, to the preference attribute set.

[0077] The preference analysis model processes the preference text vectors to obtain the set ,in Indicates Preference attributes, , the corresponding prediction score set , each Indicates the preference attribute If the preference attribute is binary, the prediction score is Can be a Boolean value or a probability value. If the preference attribute is multi-valued, Can be a continuous value or a discrete level. The preference attribute set contains the above set and collection .

[0078] For example, , the model can output probability distributions about these attributes, such as the set of preferred attributes The value of the prediction score indicates the intensity of preference for the preferred attribute, and the larger the value, the stronger the preference.

[0079] S103, the server determines the financial business data of the enterprise to be recommended, and generates a corporate profile of the enterprise to be recommended based on the financial business data and the pre-trained GBDT model.

[0080] Among them, financial business data at least includes the capital flow information, credit record information, historical interaction behavior with financial products, and preset enterprise scientific and technological innovation capabilities of the recommended enterprise. The historical interaction behavior with financial products can be understood as the behavior of purchasing or browsing financial products, such as multiple purchases, multiple clicks on the website platform to browse the relevant introduction of the financial product, etc., which can be used to characterize the preferences of enterprise users for financial products. The preset enterprise scientific and technological innovation capabilities can be an evaluation result based on the current status of the enterprise that can characterize the enterprise's scientific and technological innovation level and future development. The preset enterprise scientific and technological innovation capabilities can come from a third-party evaluation platform, and this application does not make specific restrictions on this.

[0081] In the embodiment of the present application, the financial business data can be the financial business data proactively provided by the recommended enterprise, or obtained from the enterprise database or enterprise data blockchain platform. In addition, for newly established enterprises, the server can provide enterprise portrait tags for enterprise users to select and obtain enterprise portraits.

[0082] In the embodiment of the present application, the enterprise portrait of the enterprise to be recommended is generated based on the financial business data and the pre-trained GBDT model, specifically including:

[0083] The server generates a business data matrix corresponding to the financial business data according to the data parameter type. The rows of the business data matrix correspond to different data parameter types, and the columns correspond to different times. The business data matrix is ​​input into the GBDT model for iterative calculation until the preset number of iterations is reached to obtain the portrait indicators of multiple enterprise labels corresponding to the financial business data. Enterprise labels include at least one or more of the following: enterprise type, development stage, and risk level. Based on the portrait indicators, the enterprise portrait corresponding to the recommended enterprise is constructed.

[0084] In other words, the server can use the behavior data parameter type as time to generate a business data matrix of the changes in financial business data of the enterprise during the development process at different times. That is, the business data matrix can characterize the development stage of the enterprise and the changes in the current status of the enterprise. The pre-trained GBDT (Gradient Boosting Decision Tree) is trained on a data set containing several corporate financial business data and their corresponding labels, such as enterprise type, development stage, and risk level. The server inputs the business data matrix into the GBDT model. The GBDT model undergoes multiple rounds of iterative operations, generates a weak learner for each iteration, and adjusts the weight according to the prediction error of the previous learner until the preset number of iterations is reached. The preset number of iterations is set by the user according to the actual situation, and this application does not make specific restrictions on this. Finally, the prediction results of all learners are weighted and summed to obtain the final prediction result, that is, the portrait index predicted for each enterprise label of the financial business data. The portrait index can be a predicted value, such as a risk score, or a classification label, such as a specific enterprise type. Finally, the server will generate a corporate portrait such as {technology-based, growth stage, healthy financial status...} based on the portrait index.

[0085] By constructing a corporate portrait, relevant information about the company can be displayed in a comprehensive and detailed manner, providing a basis for subsequent recommendation services and decision support.

[0086] S104, the server matches the historical product recommendation matrix of the historically recommended enterprise set corresponding to the enterprise portrait based on the preset portrait matching rules.

[0087] Among them, the preset portrait matching rules are used to compare the enterprise portrait with the corresponding recommended enterprise portraits of each historically recommended enterprise.

[0088] In the embodiment of the present application, based on the preset portrait matching rules, the historical product recommendation matrix of the historical recommended enterprise set corresponding to the enterprise portrait is matched, specifically including:

[0089] Calculate the similarity between the enterprise portrait and the recommended enterprise portraits of each historically recommended enterprise in the preset database. The portrait similarity is calculated based on the feature vectors corresponding to the enterprise portrait and the recommended enterprise portrait. Add each historically recommended enterprise whose portrait similarity is greater than the preset similarity threshold to the historically recommended enterprise set. Calculate the historical product recommendation matrix corresponding to the enterprise portrait based on the historically recommended enterprise set, the portrait similarity and the preset initial historical product recommendation matrix. The rows of the historical product recommendation matrix are each historically recommended enterprise, which is listed as the recommended historical products.

[0090] That is to say, the server calculates the similarity through the feature vector corresponding to the enterprise portrait and the feature vector of the recommended enterprise portrait of each historically recommended enterprise in the preset database. The portrait similarity calculation formula is as follows:

[0091]

[0092] in, Represents corporate image and The similarity of the portraits between the historically recommended companies, Indicates the first The image indicator component, Indicates The recommended enterprise portrait corresponding to the historical recommended enterprise The image indicator component, Indicates the number of image indicator components.

[0093] The server compares the similarity of each image obtained with the preset similarity threshold Compare and set similarity threshold It is a threshold preset by the user for filtering historical recommended companies with high similarity. The server filters profiles with similarity greater than the preset similarity threshold. The historical recommended enterprises are selected, and the financial business data of the selected historical recommended enterprises are highly similar to those of the recommended enterprises. The server adds the selected historical recommended enterprises to the historical recommended enterprise set, and obtains SRES ={ }, For the Historically recommended companies. Among them, HRES ={ }, The number of all historical recommended companies in the preset database.

[0094] Subsequently, the server generates a historical product recommendation matrix based on the historical recommended enterprise set, portrait similarity, and the preset initial historical product recommendation matrix, recommends enterprises according to behavioral history, lists historical products as recommendations, and uses portrait similarity as weight. The initial historical product recommendation matrix contains the recommendation scores of historical recommended enterprises for each financial product. The historical product recommendation matrix can be generated using the following formula:

[0095]

[0096] in, Represents the first Line Column elements, Indicates Historically recommended companies for each financial product Recommendation rating. Indicates other situations. Based on The recommendation weight after normalization. Include x The historical recommended enterprises correspond to y Recommendation ratings for financial products.

[0097] Through the above scheme, information can be extracted from the historical recommended enterprise set, and a historical product recommendation matrix for enterprise portraits can be constructed, relying on effective similarity calculation methods and reasonable recommendation weight adjustment strategies to make subsequent accurate product recommendations.

[0098] S105, the server generates a corresponding set of recommended difference products and its corresponding recommended difference preference attribute tree according to the historical product recommendation matrix and the preset financial product feature matrix.

[0099] Among them, the behaviors of different financial products in the preset financial product characteristic matrix are listed as different product characteristics; the characteristic dimensions of product characteristics at least include type, risk level, return, term, and issuing institution corresponding to each preference attribute.

[0100] In the embodiment of the present application, according to the historical product recommendation matrix and the preset financial product feature matrix, a corresponding recommended difference product set and its corresponding recommended difference preference attribute tree are generated, specifically including:

[0101] According to the historical product recommendation matrix and the preset financial product feature matrix, calculate the recommended feature values ​​of each historical product recommended by each historical recommendation enterprise. According to the recommended feature values ​​corresponding to each historical product and the historical product recommendation matrix, determine the recommended average feature vector corresponding to each historical product. According to the vector difference between the preset financial product feature matrix and each recommended average feature vector, determine the corresponding recommendation difference matrix. Compare each element in the recommendation difference matrix with the preset recommendation threshold to filter the historical products that meet the preset conditions according to the comparison results, and add them to the recommended difference product set. Use each product feature corresponding to the recommended difference product set as an attribute tree node to construct the corresponding recommendation difference preference attribute tree.

[0102] In other words, the server recommends products based on the historical product matrix And preset financial product feature matrix F , calculate the recommended feature value, , F , Represents the number of product features of financial products, and then uses each recommended feature value and the historical product recommendation matrix to calculate the recommended average feature vector. The recommended average feature vector calculation is as follows:

[0103]

[0104] in, Indicates The recommended average feature vector of financial products, Indicates The corresponding scores of the financial product characteristics of the historically recommended enterprises, Indicates The server calculates the vector difference between the preset financial product feature matrix and each recommended average vector to obtain the recommendation difference matrix. The calculation formula is as follows:

[0105]

[0106] in, Indicates The first financial product The server compares each element in the recommendation difference matrix with the preset recommendation threshold. The preset recommendation threshold can be set by the user according to actual use, or it can be set and updated based on historical experience. This application does not make specific restrictions on this. According to the comparison results, each element greater than the preset recommendation threshold is screened out, and the historical products corresponding to each screened element are determined to generate a set of recommended difference products.

[0107] Subsequently, the server can use a graph or tree structure to represent the recommendation difference preference attribute tree. When constructing it, it can arrange the product features as attribute tree nodes in sequence according to the hierarchical relationship. For example, the recommendation difference preference attribute tree contains three first-level attribute tree nodes (top-level nodes), namely: investment risk, investment period, and investment field; investment risk contains three second-level attribute tree nodes: low risk, medium risk, and high risk; low risk contains a third-level attribute tree node: product A1, medium risk contains two third-level attribute tree nodes: product A2 and product A3, and high risk contains a third-level attribute tree node: product A4... Specific preference attributes will be expanded under each top-level node, and the difference products will be associated with the corresponding nodes, so as to accurately understand user preferences and conveniently provide personalized recommendation services.

[0108] S106, the server determines the preference attribute grouping of the enterprise to be recommended based on the preference attribute set, the recommendation difference preference attribute tree and the preset preference matching degree recognition model, and matches each financial product to be recommended in the financial product library based on the preference attribute grouping and the enterprise portrait, so as to obtain a recommended financial product sequence and send it to the enterprise terminal.

[0109] In the embodiment of the present application, based on the preference attribute set, the recommendation difference preference attribute tree and the preset preference matching degree recognition model, the preference attribute grouping of the enterprise to be recommended is determined, specifically including:

[0110] Through the preset preference matching recognition model, each preference attribute in the preference attribute set is traversed through the recommended difference preference attribute tree one by one to determine the matching attribute tree node of each preference attribute. The matching degree corresponding to the matching attribute tree node is greater than the preset preference matching threshold. According to the matching attribute tree node, the preference attributes are divided into preference attribute groups, and the preference attributes are removed from the preference attribute set until the preference attribute set is an empty set.

[0111] The preset preference matching recognition model may be a preset function that can calculate the matching degree between two preference attributes and output a value between 0 and 1, where 0 indicates a complete mismatch and 1 indicates a complete match. The function may obtain the matching degree by calculating the cosine similarity or the inverse of the Euclidean distance.

[0112] That is to say, the present application uses the above-mentioned preset preference matching degree identification model to traverse each preference attribute in the preference attribute set, and calculates the matching degree of each preference attribute with each node of the above-mentioned recommended difference preference attribute tree, so as to find a matching attribute tree node, that is, the calculated matching degree of the two is greater than the preset preference matching degree threshold. The preset preference matching degree threshold can be set according to actual use, and the present application does not make specific restrictions on this. After obtaining the matching attribute tree node, the server will divide each preference attribute in the preference attribute set into the preference attribute grouping according to the preset grouping corresponding to each matching attribute tree node, thereby grouping the preference attributes in a more refined manner, ensuring that the preference attribute grouping reflects the natural classification of the preference attributes.

[0113] Preference attribute grouping is the process of organizing the preference attributes of the recommended companies according to their similarities and positions in the recommendation difference preference attribute tree. Each grouping represents a set of preferences that are logically or attribute-similar, which helps the recommendation system to more accurately understand the preferences of the recommended companies and provide more personalized and accurate recommendations for them. For example, one grouping may contain all the preference attributes related to "low-risk investment", while another grouping may contain attributes related to "high return". Through such grouping, the recommendation system can more easily find financial products that match the preferences of the recommended companies.

[0114] In addition, if there are still preference attributes in the preference attribute set, and the preference attributes have not been classified into any preference attribute group, then the preference attributes can be identified as personalized preference attributes of the enterprise to be recommended. The server adds these part of the preference attributes to a preference attribute group with a higher priority to give priority to displaying the financial products to be recommended that match the preference attribute group.

[0115] In the embodiment of the present application, before matching each to-be-recommended financial product in the financial product library based on the preference attribute grouping and the enterprise portrait to obtain a recommended financial product sequence and sending it to the enterprise terminal, the following is also included:

[0116] The server determines the corresponding similar evaluation information in the preset evaluation list according to the enterprise profile. The preset evaluation list includes the scores of several enterprises on each financial product to be recommended. The similar evaluation information includes the scores of each financial product to be recommended by the enterprise corresponding to the enterprise profile. The evaluation similarity between each financial product to be recommended is calculated according to the similar evaluation information, the preset evaluation list and the preset collaborative filtering algorithm. The recommendation score corresponding to each financial product to be recommended is determined according to each evaluation similarity, so that each financial product to be recommended with a recommendation score greater than a score threshold is regarded as a financial product to be recommended that matches the enterprise profile.

[0117] That is, the server can search for similar evaluation information that matches the enterprise portrait from the preset evaluation list through the enterprise portrait. The similar evaluation information that matches the enterprise portrait can be understood as the ratings of the recommended enterprise or enterprises similar to the recommended enterprise on the recommended financial product. The server uses the preset collaborative filtering algorithm, similar evaluation information and the preset evaluation list to calculate the evaluation similarity between each financial product to be recommended, and the calculation is as follows:

[0118]

[0119] in, Financial products to be recommended and The similarity of the evaluations between and With different values, covering all financial products to be recommended; An object (enterprise) with similar evaluation information that represents the enterprise portrait matching. An object representing multiple similar evaluation information. Representation Object Recommended financial products Rating, Representation Object The average score of all financial products to be recommended. Representation Object Recommended financial products Rating, Indicates the attitude towards recommending financial products The number of objects that have been evaluated, Indicates the attitude towards recommending financial products The number of objects that have been evaluated.

[0120] The server calculates the recommendation score for each financial product to be recommended based on the above evaluation similarity. The specific calculation formula is as follows:

[0121]

[0122] in, Indicates the recommended companies' attitude towards recommended financial products A recommendation rating; Indicates that the object is and The collection to be evaluated; Indicates that each object is simultaneously and When evaluating, the similar evaluation information and The mean and the average of the score differences (i.e. the sum of the means of the score differences); Recommend financial products for Ratings; express and The number of times it occurs simultaneously.

[0123] Through the above calculation formula, not only can the recommendation score of the recommended financial products be obtained, but also the factor that the recommended financial products appear in pairs can be taken into consideration, thereby strengthening the correlation between the recommended financial products, so that when recommending financial products for the recommended enterprises, financial products with high similarity can be associated and recommended, while financial products with low similarity can be ignored. This can achieve efficient and accurate financial product recommendations.

[0124] In addition, different financial products in this application can be actively labeled and updated by financial institutions. When financial products are listed on the system corresponding to this application, the product labels can be used to perform an association operation with the corporate portrait. The association between the product label and the corporate portrait can be related to the recommendation score. For example, when the recommendation score is higher than a preset score value, a list of associations is established between the product label and the corporate portrait. Subsequently, the list of associations established between the product label and the corporate portrait can be used by newly established companies or other companies to recommend financial products.

[0125] In the embodiment of the present application, the financial products to be recommended in the financial product library are matched based on the preference attribute grouping and the enterprise portrait to obtain the recommended financial product sequence and send it to the enterprise terminal, which specifically includes:

[0126] According to the preference attribute grouping, the financial products to be recommended that match the corporate profile are screened to obtain recommended financial products. According to the preset sequence length, in descending order of recommendation scores, each recommended financial product is added to the recommended financial product sequence of the preset sequence length.

[0127] In other words, the server can filter out various financial products to be recommended by grouping by preference attributes, for example, filter out low-risk product A1, short-term investment product B1, and technology investment product C1, and obtain recommended financial products A1, B1, and C1. The server can sort the recommended financial products from large to small according to the recommended score according to the preset sequence length, and generate a recommended financial product sequence.

[0128] In addition, the present application can also generate a number of display pages corresponding to the preference attribute groups according to preference attribute groups, and different display pages display recommended financial product sequences corresponding to different preference attribute groups, so that corporate users can view the recommended financial product sequences corresponding to different preference attribute groups in pages.

[0129] Through the above scheme, this application can deeply analyze the preference description text entered by the enterprise, and combine the preference analysis model to obtain a set of preference attributes corresponding to the personalized needs of the enterprise. And using the GBDT model to analyze the enterprise portrait, it is helpful to deeply understand the actual situation of the enterprise in order to accurately recommend financial products. Using the preset portrait matching rules, it is possible to quickly find historical recommendation cases with high similarity, and further obtain a recommendation difference preference attribute tree that can be targeted and differentiated. Based on the preference attribute set, the recommendation difference preference attribute tree and the preference matching degree identification model, the enterprise preference attributes are grouped to match the financial products to be recommended, thereby achieving personalized recommendations, significantly improving the user experience of the enterprise, enhancing user stickiness, and promoting the effective promotion of financial products.

[0130] In addition, this application can help enterprises quickly find financial products that suit their needs, guide financial institutions to optimize resource allocation, and help financial institutions better understand market demand and develop financial products that better meet market demand. This will overcome the current problems that financial products cannot be effectively and accurately recommended to corporate users, affecting the system experience of corporate users, and that financial product resources cannot be reasonably utilized.

[0131] Figure 2 The embodiment of the present application provides a personalized recommendation system 200 for corporate financial products based on preference analysis, such as Figure 2 As shown, the system 200 includes:

[0132] The first determination module 201 is used to determine the corresponding enterprise preference description text group based on the enterprise input information from the enterprise terminal. The second determination module 202 is used to determine the preference attribute set of the enterprise to be recommended based on the enterprise preference description text group and the preset preference analysis model. The determination generation module 203 is used to determine the financial business data of the enterprise to be recommended, and generate the enterprise portrait of the enterprise to be recommended based on the financial business data and the pre-trained GBDT model. Among them, the financial business data at least includes the capital flow information, credit record information, historical interaction behavior with financial products, and preset enterprise scientific and technological innovation capabilities of the enterprise to be recommended. The matching module 204 is used to match the historical product recommendation matrix of the historical recommended enterprise set corresponding to the enterprise portrait based on the preset portrait matching rules. Among them, the preset portrait matching rules are used to compare the enterprise portrait with the corresponding recommended enterprise portrait of each historically recommended enterprise. The generation module 205 is used to generate the corresponding recommended difference product set and its corresponding recommended difference preference attribute tree based on the historical product recommendation matrix and the preset financial product feature matrix. The third determination module 206 is used to determine the preference attribute grouping of the enterprise to be recommended based on the preference attribute set, the recommendation difference preference attribute tree and the preset preference matching degree recognition model, so as to match each financial product to be recommended in the financial product library based on the preference attribute grouping and the enterprise portrait, so as to obtain a recommended financial product sequence and send it to the enterprise terminal.

[0133] In the embodiment of the present application, the first determining module 201 is specifically used for:

[0134] The enterprise input information from the enterprise terminal is input into the natural language processing model to perform word segmentation processing on the enterprise input information to obtain multiple pending preference description texts. Each pending preference description text is encoded to generate a corresponding pending preference description text vector. The product between the pending preference description text vector and the first vector and the second vector is calculated to compare the corresponding first product value and the second product value with the preset completeness evaluation standard. Among them, the first vector is obtained based on several basic preference description keywords. The second vector is obtained based on several specific preference description keywords. According to the comparison result, the completeness of the preference description corresponding to the enterprise input information is determined. When the completeness of the preference description is greater than a predetermined value, each pending preference description text is added to the enterprise preference description text group. Otherwise, based on the comparison result, the corresponding basic preference dimension and / or specific preference dimension input supplementary prompt information is generated, and the input supplementary prompt information is sent to the enterprise terminal.

[0135] In the embodiment of the present application, the second determining module 202 is specifically used for:

[0136] Determine the preference text vector corresponding to the enterprise preference description text group. Input the preference text vector into the preference analysis model to determine the corresponding preference attributes and their corresponding prediction scores. The preference analysis model is a pre-trained logistic regression model for identifying preference attributes. Add each preference attribute and its corresponding prediction score to the preference attribute set.

[0137] In the embodiment of the present application, the determination generation module 203 is specifically used for:

[0138] According to the data parameter type, a business data matrix corresponding to the financial business data is generated. The rows of the business data matrix correspond to different data parameter types, and the columns correspond to different times. The business data matrix is ​​input into the GBDT model for iterative calculation until the preset number of iterations is reached to obtain the portrait indicators of multiple enterprise labels corresponding to the financial business data. Enterprise labels include at least one or more of the following: enterprise type, development stage, and risk level. According to the portrait indicators, the enterprise portrait corresponding to the recommended enterprise is constructed.

[0139] In the embodiment of the present application, the matching module 204 is specifically used for:

[0140] Calculate the similarity between the enterprise portrait and the recommended enterprise portraits of each historically recommended enterprise in the preset database. The portrait similarity is calculated based on the feature vectors corresponding to the enterprise portrait and the recommended enterprise portrait. Add each historically recommended enterprise whose portrait similarity is greater than the preset similarity threshold to the historically recommended enterprise set. Calculate the historical product recommendation matrix corresponding to the enterprise portrait based on the historically recommended enterprise set, the portrait similarity and the preset initial historical product recommendation matrix. The rows of the historical product recommendation matrix are each historically recommended enterprise, which is listed as the recommended historical products.

[0141] In the embodiment of the present application, the generation module 205 is specifically used for:

[0142] According to the historical product recommendation matrix and the preset financial product feature matrix, calculate the recommended feature values ​​of each historical product recommended by each historical recommendation enterprise. According to the recommended feature values ​​corresponding to each historical product and the historical product recommendation matrix, determine the recommended average feature vector corresponding to each historical product. According to the vector difference between the preset financial product feature matrix and each recommended average feature vector, determine the corresponding recommendation difference matrix. Compare each element in the recommendation difference matrix with the preset recommendation threshold to filter the historical products that meet the preset conditions according to the comparison results, and add them to the recommended difference product set. Use each product feature corresponding to the recommended difference product set as an attribute tree node to construct the corresponding recommendation difference preference attribute tree.

[0143] In the embodiment of the present application, the third determination module 206 is specifically used to:

[0144] Through the preset preference matching recognition model, each preference attribute in the preference attribute set is traversed through the recommended difference preference attribute tree one by one to determine the matching attribute tree node of each preference attribute. The matching degree corresponding to the matching attribute tree node is greater than the preset preference matching threshold. According to the matching attribute tree node, the preference attributes are divided into preference attribute groups, and the preference attributes are removed from the preference attribute set until the preference attribute set is an empty set.

[0145] In the embodiment of the present application, the system can also:

[0146] According to the enterprise profile, the corresponding similar evaluation information in the preset evaluation list is determined. The preset evaluation list includes the scores of several enterprises on each financial product to be recommended. The similar evaluation information includes the scores of each financial product to be recommended by the enterprise corresponding to the enterprise profile. According to the similar evaluation information, the preset evaluation list and the preset collaborative filtering algorithm, the evaluation similarity between each financial product to be recommended is calculated. According to each evaluation similarity, the recommendation score corresponding to each financial product to be recommended is determined, so that each financial product to be recommended with a recommendation score greater than the score threshold is regarded as a financial product to be recommended that matches the enterprise profile.

[0147] In the embodiment of the present application, the third determination module 206 is specifically used to:

[0148] According to the preference attribute grouping, the financial products to be recommended that match the corporate profile are screened to obtain recommended financial products. According to the preset sequence length, in descending order of recommendation scores, each recommended financial product is added to the recommended financial product sequence of the preset sequence length.

[0149] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0150] The system and method provided in the embodiment of the present application correspond one to one, and therefore the system also has similar beneficial technical effects as its corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the system will not be repeated here.

[0151] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0152] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for personalized recommendation of corporate financial products based on preference analysis, characterized in that: The method comprises: Determine a corresponding enterprise preference description text group based on enterprise input information from an enterprise terminal; Determine the preference attribute set of the enterprise to be recommended according to the enterprise preference description text group and the preset preference analysis model; Determine the financial business data of the enterprise to be recommended, and generate a corporate profile of the enterprise to be recommended based on the financial business data and the pre-trained GBDT model; wherein the financial business data at least includes the capital flow information, credit record information, historical interaction behavior with financial products, and preset corporate scientific and technological innovation capabilities of the enterprise to be recommended; Based on the preset portrait matching rules, the historical product recommendation matrix of the historical recommended enterprise set corresponding to the enterprise portrait is matched; wherein the preset portrait matching rules are used to compare the enterprise portrait with the recommended enterprise portrait corresponding to each historically recommended enterprise; Generate a corresponding recommended difference product set and its corresponding recommended difference preference attribute tree according to the historical product recommendation matrix and the preset financial product feature matrix; Based on the preference attribute set, the recommendation difference preference attribute tree and the preset preference matching degree recognition model, the preference attribute grouping of the enterprise to be recommended is determined, so as to match each financial product to be recommended in the financial product library based on the preference attribute grouping and the enterprise portrait, so as to obtain a recommended financial product sequence and send it to the enterprise terminal; Wherein, based on the enterprise input information from the enterprise terminal, the corresponding enterprise preference description text group is determined, specifically including: Inputting the enterprise input information from the enterprise terminal into a natural language processing model to perform word segmentation processing on the enterprise input information to obtain a plurality of pending preference description texts; Encoding each of the pending preference description texts to generate a corresponding pending preference description text vector; Calculating the product of the pending preference description text vector with the first vector and the second vector respectively, so as to compare the corresponding first product value and the second product value with the preset completeness evaluation standard respectively; wherein the first vector is obtained based on a number of basic preference description keywords; and the second vector is obtained based on a number of specific preference description keywords; According to the comparison results, determine the completeness of the preference description corresponding to the enterprise input information; When the completeness of the preference description is greater than a predetermined value, each of the pending preference description texts is added to the enterprise preference description text group; Otherwise, based on the comparison result, generate corresponding supplementary prompt information for entry of the basic preference dimension and / or the specific preference dimension, and send the supplementary prompt information for entry to the enterprise terminal; Wherein, based on the preset portrait matching rules, the historical product recommendation matrix of the historical recommended enterprise set corresponding to the enterprise portrait is matched, specifically including: Calculating the similarity between the enterprise portrait and the recommended enterprise portraits of each of the historically recommended enterprises in a preset database; wherein the similarity is calculated based on the feature vectors corresponding to the enterprise portrait and the recommended enterprise portrait respectively; Adding each of the historically recommended enterprises whose portrait similarity is greater than a preset similarity threshold to the historically recommended enterprise set; According to the historical recommended enterprise set, the portrait similarity and the preset initial historical product recommendation matrix, the historical product recommendation matrix corresponding to the enterprise portrait is calculated; the rows of the historical product recommendation matrix are each historically recommended enterprise, which is listed as each recommended historical product; According to the historical product recommendation matrix and the preset financial product feature matrix, a corresponding recommended difference product set and its corresponding recommended difference preference attribute tree are generated, specifically including: Calculate the recommended feature values ​​of each of the historical products recommended by each of the historically recommended enterprises according to the historical product recommendation matrix and the preset financial product feature matrix; Determining a recommended average feature vector corresponding to each of the historical products according to the recommended feature value corresponding to each of the historical products and the historical product recommendation matrix; Determine a corresponding recommendation difference matrix according to the vector differences between the preset financial product feature matrix and each of the recommended average feature vectors; Compare each element in the recommendation difference matrix with a preset recommendation threshold value, so as to filter the historical products that meet the preset conditions according to the comparison results, and add them to the recommended difference product set; Taking each product feature corresponding to the recommended difference product set as an attribute tree node, constructing a corresponding recommended difference preference attribute tree; Wherein, based on the preference attribute set, the recommendation difference preference attribute tree and the preset preference matching degree recognition model, determining the preference attribute grouping of the to-be-recommended enterprise specifically includes: Through the preset preference matching degree recognition model, each preference attribute in the preference attribute set is traversed through the recommendation difference preference attribute tree one by one to determine a matching attribute tree node of each preference attribute; the matching degree corresponding to the matching attribute tree node is greater than a preset preference matching degree threshold; According to the matching attribute tree nodes, the preference attributes are divided into the preference attribute groups, and the preference attributes are removed from the preference attribute set until the preference attribute set is an empty set; Wherein, before matching each to-be-recommended financial product in the financial product library based on the preference attribute grouping and the enterprise portrait to obtain a recommended financial product sequence and sending it to the enterprise terminal, the method further includes: Determine the corresponding similar evaluation information in a preset evaluation list according to the enterprise portrait; wherein the preset evaluation list includes scores of several enterprises on each of the financial products to be recommended; and the similar evaluation information includes scores of the enterprises corresponding to the enterprise portrait on each of the financial products to be recommended; Calculating the evaluation similarity between the financial products to be recommended according to the similar evaluation information, the preset evaluation list and the preset collaborative filtering algorithm; According to the similarities of each of the evaluations, a recommendation score corresponding to each of the financial products to be recommended is determined, so that each of the financial products to be recommended whose recommendation score is greater than a score threshold is regarded as the financial product to be recommended that matches the corporate profile.

2. The method for personalized recommendation of corporate financial products based on preference analysis according to claim 1, characterized in that: According to the enterprise preference description text group and the preset preference analysis model, a set of preference attributes of the enterprise to be recommended is determined, specifically including: Determine a preference text vector corresponding to the enterprise preference description text group; Inputting the preference text vector into the preference analysis model to determine the corresponding preference attributes and their corresponding prediction scores; wherein the preference analysis model is a pre-trained logistic regression model for identifying preference attributes; Each of the preference attributes and the corresponding prediction score are added to the preference attribute set.

3. The method for personalized recommendation of corporate financial products based on preference analysis according to claim 1, characterized in that: Generate the enterprise profile of the enterprise to be recommended based on the financial business data and the pre-trained GBDT model, specifically including: Generate a business data matrix corresponding to the financial business data according to the data parameter type; the rows of the business data matrix correspond to different data parameter types, and the columns correspond to different times; Input the business data matrix into the GBDT model to perform iterative operations until a preset number of iterations is reached, and obtain the portrait indicators of multiple enterprise labels corresponding to the financial business data; the enterprise labels include at least one or more of the following: enterprise type, development stage, risk level; According to the portrait indicators, a corporate portrait corresponding to the to-be-recommended enterprise is constructed.

4. The method for personalized recommendation of corporate financial products based on preference analysis according to claim 1, characterized in that: Based on the preference attribute grouping and the enterprise portrait, each financial product to be recommended in the financial product library is matched to obtain a recommended financial product sequence and send it to the enterprise terminal, specifically including: According to the preference attribute grouping, the financial products to be recommended that match the corporate profile are screened to obtain recommended financial products; According to the preset sequence length, each of the recommended financial products is sequentially added to the recommended financial product sequence of the preset sequence length in descending order of the recommendation score.

5. A personalized recommendation system for corporate financial products based on preference analysis, characterized in that: The system comprises: A first determination module, used to determine a corresponding enterprise preference description text group based on enterprise input information from an enterprise terminal; A second determination module is used to determine a set of preference attributes of the enterprise to be recommended according to the enterprise preference description text group and a preset preference analysis model; A determination generation module is used to determine the financial business data of the enterprise to be recommended, and generate a corporate portrait of the enterprise to be recommended based on the financial business data and the pre-trained GBDT model; wherein the financial business data at least includes the capital flow information, credit record information, historical interaction behavior with financial products, and preset corporate scientific and technological innovation capabilities of the enterprise to be recommended; A matching module, for matching the historical product recommendation matrix of the historically recommended enterprise set corresponding to the enterprise portrait based on a preset portrait matching rule; wherein the preset portrait matching rule is used to compare the enterprise portrait with the recommended enterprise portrait corresponding to each historically recommended enterprise; A generation module, used to generate a corresponding set of recommended difference products and a corresponding recommended difference preference attribute tree according to the historical product recommendation matrix and the preset financial product feature matrix; A third determination module is used to determine the preference attribute grouping of the enterprise to be recommended based on the preference attribute set, the recommendation difference preference attribute tree and the preset preference matching degree recognition model, so as to match each financial product to be recommended in the financial product library based on the preference attribute grouping and the enterprise portrait, so as to obtain a recommended financial product sequence and send it to the enterprise terminal; Wherein, the first determining module is specifically used for: Inputting the enterprise input information from the enterprise terminal into a natural language processing model to perform word segmentation processing on the enterprise input information to obtain a plurality of pending preference description texts; Encoding each of the pending preference description texts to generate a corresponding pending preference description text vector; Calculating the product of the pending preference description text vector with the first vector and the second vector respectively, so as to compare the corresponding first product value and the second product value with the preset completeness evaluation standard respectively; wherein the first vector is obtained based on a number of basic preference description keywords; and the second vector is obtained based on a number of specific preference description keywords; According to the comparison results, determine the completeness of the preference description corresponding to the enterprise input information; When the completeness of the preference description is greater than a predetermined value, each of the pending preference description texts is added to the enterprise preference description text group; Otherwise, based on the comparison result, generate corresponding supplementary prompt information for entry of the basic preference dimension and / or the specific preference dimension, and send the supplementary prompt information for entry to the enterprise terminal; Wherein, the matching module is specifically used for: Calculating the similarity between the enterprise portrait and the recommended enterprise portraits of each of the historically recommended enterprises in a preset database; wherein the similarity is calculated based on the feature vectors corresponding to the enterprise portrait and the recommended enterprise portrait respectively; Adding each of the historically recommended enterprises whose portrait similarity is greater than a preset similarity threshold to the historically recommended enterprise set; According to the historical recommended enterprise set, the portrait similarity and the preset initial historical product recommendation matrix, the historical product recommendation matrix corresponding to the enterprise portrait is calculated; the rows of the historical product recommendation matrix are each historically recommended enterprise, which is listed as each recommended historical product; Wherein, the generation module is specifically used for: Calculate the recommended feature values ​​of each of the historical products recommended by each of the historically recommended enterprises according to the historical product recommendation matrix and the preset financial product feature matrix; Determining a recommended average feature vector corresponding to each of the historical products according to the recommended feature value corresponding to each of the historical products and the historical product recommendation matrix; Determine a corresponding recommendation difference matrix according to the vector differences between the preset financial product feature matrix and each of the recommended average feature vectors; Compare each element in the recommendation difference matrix with a preset recommendation threshold value, so as to filter the historical products that meet the preset conditions according to the comparison results, and add them to the recommended difference product set; Taking each product feature corresponding to the recommended difference product set as an attribute tree node, constructing a corresponding recommended difference preference attribute tree; Wherein, the third determination module is specifically used for: Through the preset preference matching degree recognition model, each preference attribute in the preference attribute set is traversed through the recommendation difference preference attribute tree one by one to determine a matching attribute tree node of each preference attribute; the matching degree corresponding to the matching attribute tree node is greater than a preset preference matching degree threshold; According to the matching attribute tree nodes, the preference attributes are divided into the preference attribute groups, and the preference attributes are removed from the preference attribute set until the preference attribute set is an empty set; Among other things, the system can also: Determine the corresponding similar evaluation information in a preset evaluation list according to the enterprise portrait; wherein the preset evaluation list includes scores of several enterprises on each of the financial products to be recommended; and the similar evaluation information includes scores of the enterprises corresponding to the enterprise portrait on each of the financial products to be recommended; Calculating the evaluation similarity between the financial products to be recommended according to the similar evaluation information, the preset evaluation list and the preset collaborative filtering algorithm; According to the similarities of each of the evaluations, a recommendation score corresponding to each of the financial products to be recommended is determined, so that each of the financial products to be recommended whose recommendation score is greater than a score threshold is regarded as the financial product to be recommended that matches the corporate profile.

Citation Information

Patent Citations

  • Policy information pushing method and device

    CN113421172A

  • Financial product pushing method and device, equipment, medium and computer program product

    CN114170026A