Security information intelligent screening method and system based on natural language processing

Abstract

The invention relates to the technical field of financial information, and discloses an intelligent security information screening method and system based on natural language processing, and the method comprises the steps: analyzing a query demand intention and user feature information corresponding to a service user, and analyzing the browsing behavior preference of the service user; constructing an information recommendation tendency of the service user, obtaining a security information database of the security system, analyzing security information elements corresponding to the security information database, and calculating the integrating degree between the information recommendation tendency and the security information elements; scheduling historical security trade data of the service user in the security system, analyzing security trade behavior characteristics of the service user, and performing information filtering processing on the initial security information based on the security trade behavior characteristics to obtain filtered security information; and evaluating the risk tolerance corresponding to the service user, and performing optimization processing on the filtered security information to obtain screened security information of the service user. According to the invention, the security information screening accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to a method and system for intelligent screening of securities information based on natural language processing, and belongs to the field of financial information technology. Background Art

[0002] Securities information refers to various information and materials related to the securities market, securities products, and related financial activities. In order to improve the accuracy of users' investment decisions and increase user attractiveness, securities institutions will recommend personalized securities information to users to help users understand the dynamic information of the securities market. Therefore, securities information needs to be screened and processed.

[0003] The current existing securities information screening method adopts manual screening and keyword matching methods. Professional financial analysts mainly use their own professional knowledge and experience to analyze and screen securities information, combine the screened information with the user's query content, and perform a second screening of the screened information through keyword matching to finally obtain the target information. However, this method does not take the user's personal behavior habits into consideration during the processing process, resulting in a low matching degree of the screened information. In addition, the screened information does not take into account the corresponding investment risks, which in turn reduces the accuracy of securities information screening. Summary of the Invention

[0004] The present invention provides a method and system for intelligent screening of securities information based on natural language processing, the main purpose of which is to improve the accuracy of securities information screening.

[0005] To achieve the above objectives, the present invention provides a method for intelligently screening securities information based on natural language processing, comprising:

[0006] Utilizing the securities system to receive a request instruction from a service user, parsing the query demand intention and user characteristic information corresponding to the service user based on the request instruction, and analyzing the browsing behavior preference of the service user based on the user characteristic information;

[0007] Combining the browsing behavior preference and the query demand intention, constructing the information recommendation tendency of the service user, obtaining the securities information database of the securities system, parsing the securities information elements corresponding to the securities information database, and calculating the degree of fit between the information recommendation tendency and the securities information elements;

[0008] When the degree of compatibility is greater than a preset degree of compatibility, initial securities information is screened from the securities information database, historical securities trading data of the service user is dispatched in the securities system, securities trading behavior characteristics of the service user are analyzed based on the historical securities trading data, and information filtering processing is performed on the initial securities information based on the securities trading behavior characteristics to obtain filtered securities information;

[0009] The user asset information corresponding to the service user is queried, and based on the user asset information, the risk tolerance corresponding to the service user is evaluated; based on the risk tolerance, the filtered securities information is optimized to obtain the filtered securities information of the service user.

[0010] Optionally, parsing the query demand intention and user feature information corresponding to the service user based on the request instruction includes:

[0011] Performing protocol parsing on the request instruction to obtain a request instruction packet;

[0012] Performing word sense disambiguation on the request instruction packet to obtain a request instruction text;

[0013] Performing intent classification on the request instruction text to obtain a classified instruction text;

[0014] Extracting parameters from the classified instruction text to obtain instruction text parameters;

[0015] Based on the instruction text parameters, the query demand intention and user characteristic information corresponding to the service user are analyzed.

[0016] Optionally, analyzing the browsing behavior preferences of the service user based on the user characteristic information includes:

[0017] Determining user identity information corresponding to the service user based on the user characteristic information;

[0018] Based on the user identity information, obtaining cross-platform browsing data corresponding to the service user;

[0019] Extracting multidimensional browsing features from the cross-platform browsing data, performing cluster analysis on the multidimensional browsing features, and obtaining clustered browsing features;

[0020] Performing behavior preference analysis on the clustered browsing features to obtain the browsing behavior preferences of the service users.

[0021] Optionally, combining the browsing behavior preference and the query demand intention to construct the information recommendation tendency of the service user includes:

[0022] Extracting semantic features of the browsing behavior preference to obtain a behavior semantic feature vector;

[0023] Extracting intent keywords from the query requirement intent and constructing intent feature vectors corresponding to the intent keywords;

[0024] Performing vector fusion processing on the behavior semantic feature vector and the intention feature vector to obtain a comprehensive feature vector;

[0025] Performing feature enhancement processing on the comprehensive feature vector to obtain an enhanced feature vector;

[0026] performing interest mapping processing on the enhanced feature vector to obtain an interest mapping matrix;

[0027] Calculating an interest score corresponding to the service user based on the interest mapping matrix;

[0028] Based on the interest score and the interest mapping matrix, the information recommendation tendency of the service user is constructed.

[0029] Optionally, calculating the interest score corresponding to the service user based on the interest mapping matrix includes:

[0030] Calculating the correlation between each matrix element in the interest mapping matrix to obtain the element correlation;

[0031] Normalizing the interest mapping matrix to obtain a normalized interest matrix;

[0032] Performing weight assignment on each matrix element in the normalized interest matrix to obtain element weight;

[0033] Combining the element relevance and the element weight, the interest score corresponding to the service user is calculated using the following formula:

[0034]

[0035] Among them, A represents the interest score of the service user, D a，a+1 represents the element correlation between the ath element and the a+1th element in the normalized interest matrix, E a represents the element weight of the ath element in the normalized interest matrix, λ represents the time decay coefficient, a represents the element sequence number of the normalized interest matrix, and q represents the number of elements in the normalized interest matrix.

[0036] Optionally, parsing the securities information elements corresponding to the securities information database includes:

[0037] Identifying the securities information tag corresponding to the securities information database and extracting information entity data from the securities information database;

[0038] Performing attribute annotation on the information entity data to obtain information entity attributes;

[0039] Filtering out representative information tags from the securities information tags, and calculating similarity coefficients between the representative information tags and the information entity attributes;

[0040] extracting key information entities from the information entity data based on the similarity coefficient;

[0041] The securities information elements corresponding to the securities information database are parsed from the key information entities.

[0042] Optionally, the calculating the degree of compatibility between the information recommendation tendency and the securities information elements includes:

[0043] Performing vectorization processing on the information recommendation tendency and the securities information elements respectively to obtain a recommendation tendency vector and an information element vector;

[0044] Combining the recommendation tendency vector and the information element vector, the degree of fit between the information recommendation tendency and the securities information element is calculated using the following formula:

[0045]

[0046] Among them, B represents the degree of fit between information recommendation tendency and securities information elements. represents the recommendation tendency vector, represents the information element vector, The vector modulus of the recommendation tendency vector, Indicates the vector modulus of the information element vector.

[0047] Optionally, analyzing the securities trading behavior characteristics of the service user based on the historical securities trading data includes:

[0048] Performing data cleaning on the historical securities trading data to obtain target securities trading data;

[0049] Classifying and integrating the target securities trading data to obtain integrated securities trading data;

[0050] Performing basic feature statistics on the integrated securities trading data to obtain initial features of securities trading, and analyzing trading behaviors corresponding to the initial features of securities trading;

[0051] analyzing feature correlation relationships among the initial features of securities trading, and constructing a feature correlation network among the initial features of securities trading based on the feature correlation relationships;

[0052] Based on the feature association network, core trade features are screened out from the initial features of the securities trade;

[0053] The trading behavior and the core trading characteristics are combined to generate the securities trading behavior characteristics of the service user.

[0054] Optionally, the assessing the risk tolerance of the service user based on the user's financial information includes:

[0055] Verifying the user's asset information to obtain trusted asset information;

[0056] Classify the trusted asset information to obtain liability asset information and income asset information;

[0057] Extracting the debt scale and debt structure from the debt asset information, and calculating the debt stress index corresponding to the service user based on the debt scale and the debt structure;

[0058] Extracting the income scale and income composition from the income asset information, and calculating the income elasticity index corresponding to the service user based on the income scale and the income composition;

[0059] The risk tolerance of the service user is evaluated by combining the debt pressure index and the income elasticity index.

[0060] In order to solve the above problems, the present invention further provides a securities information intelligent screening system based on natural language processing, the system comprising:

[0061] a behavior preference analysis module configured to receive a request instruction from a service user using the securities system, parse the request instruction to determine the query demand intention and user characteristic information corresponding to the service user, and analyze the browsing behavior preference of the service user based on the user characteristic information;

[0062] a compatibility calculation module, configured to combine the browsing behavior preference and the query demand intention to construct the information recommendation tendency of the service user, obtain the securities information database of the securities system, parse the securities information elements corresponding to the securities information database, and calculate the compatibility between the information recommendation tendency and the securities information elements;

[0063] an information filtering module, configured to, when the degree of compatibility is greater than a preset degree of compatibility, filter initial securities information from the securities information database, dispatch historical securities trading data of the service user in the securities system, analyze securities trading behavior characteristics of the service user based on the historical securities trading data, and perform information filtering on the initial securities information based on the securities trading behavior characteristics to obtain filtered securities information;

[0064] The information optimization module is used to query the user asset information corresponding to the service user, evaluate the risk tolerance corresponding to the service user based on the user asset information, and optimize the filtered securities information based on the risk tolerance to obtain the filtered securities information of the service user.

[0065] Compared to the problems described in the background art, the present invention, by parsing the query intent and user characteristic information corresponding to the service user based on the request instruction, can convert the diverse instructions of the service user into structured information that the system can understand, thereby avoiding subsequent service deviations. Furthermore, the present invention, by combining the browsing behavior preferences and the query intent, constructs the information recommendation tendency of the service user, can deeply understand the user's personalized needs, accurately grasp the user's focus and interest in securities information, thereby achieving refined services for users and improving the accuracy and effectiveness of securities information recommendations. The present invention, by analyzing the securities trading behavior characteristics of the service user based on the historical securities trading data, can understand the investment habits and trading patterns of the service user, thereby providing an important basis for subsequent information filtering of the initial securities information. Furthermore, the present invention, by evaluating the risk tolerance corresponding to the service user based on the user's financial information, can understand the risk tolerance of the service user through the risk tolerance, thereby preventing the user from taking excessive risks due to receiving inappropriate securities information, thereby improving the effectiveness and accuracy of the information. Therefore, the intelligent securities information screening method and system based on natural language processing provided by the embodiments of the present invention can improve the accuracy of securities information screening. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A flowchart of a method for intelligent screening of securities information based on natural language processing provided by one embodiment of the present invention;

[0067] Figure 2 Schematic diagram of behavioral preference analysis of the intelligent screening method for securities information based on natural language processing provided by the present invention;

[0068] Figure 3 A schematic diagram of a module for implementing the intelligent screening method for securities information based on natural language processing provided in one embodiment of the present invention.

[0069] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0070] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0071] The present embodiment provides a method for intelligently screening securities information based on natural language processing. The method can be executed by at least one of a server, a terminal, or other electronic device capable of executing the method provided by the present embodiment. In other words, the method can be executed by software or hardware installed on a terminal or server. The server can include, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0072] Example 1:

[0073] Reference Figure 1 FIG. 1 is a flow chart of a method for intelligently screening securities information based on natural language processing according to an embodiment of the present invention. In this embodiment, the method for intelligently screening securities information based on natural language processing includes:

[0074] S1. Utilize the securities system to receive a request instruction from a service user, parse out the query demand intention and user feature information corresponding to the service user based on the request instruction, and analyze the browsing behavior preference of the service user based on the user feature information.

[0075] The present invention parses the query demand intention and user characteristic information corresponding to the service user based on the request instruction, and can convert the diverse instructions of the service user into structured information that can be understood by the system, thereby avoiding subsequent service deviations. The securities system is a platform or software system for receiving, processing and feeding back securities-related operation instructions of service users; the service users are individuals or institutional entities that conduct various securities-related activities such as investment and inquiry in the securities system; the request instruction is an instruction issued by the service user to the securities system through the client to implement specific securities business operations; the query demand intention is the core purpose that the service user wants to achieve by issuing the request instruction; and the user characteristic information is data information that can reflect the attributes, behavioral habits and investment characteristics of the service user.

[0076] As an embodiment of the present invention, parsing the query demand intention and user characteristic information corresponding to the service user based on the request instruction includes:

[0077] Performing protocol parsing on the request instruction to obtain a request instruction packet;

[0078] Performing word sense disambiguation on the request instruction packet to obtain a request instruction text;

[0079] Performing intent classification on the request instruction text to obtain a classified instruction text;

[0080] Extracting parameters from the classified instruction text to obtain instruction text parameters;

[0081] Based on the instruction text parameters, the query demand intention and user characteristic information corresponding to the service user are analyzed.

[0082] Among them, the request instruction package is the structured data formed after the protocol parsing of the request instruction; the request instruction text is the text content obtained after the word sense disambiguation processing of the request instruction package is carried out to eliminate the lexical ambiguity; the classified instruction text is the text obtained after the request instruction text is classified into different intent categories through intent classification; the instruction text parameter is the key data information obtained by extracting parameters from the classified instruction text.

[0083] Furthermore, the request instruction can be parsed through a network protocol analysis tool to obtain a request instruction package; the request instruction package can be disambiguated through a semantic disambiguation algorithm based on a knowledge graph to obtain a request instruction text; the request instruction text can be classified through an intent classification model based on deep learning to obtain a classified instruction text; the classified instruction text can be parameter extracted through a method combining regular expression matching with semantic role labeling to obtain instruction text parameters; based on the descriptive content in the instruction text parameters, the query demand intention and user feature information corresponding to the service user are analyzed, such as analyzing the query demand intention based on intent-related content, and analyzing the user feature information based on user identity information.

[0084] The present invention analyzes the browsing behavior preferences of the service users based on the user characteristic information, thereby understanding the personalized behavioral characteristics of the service users' browsing tendencies, and further providing a basis for accurately pushing securities information and services that meet the users' needs, wherein the browsing behavior preferences are expressions of the tendency choices and habits shown by the service users.

[0085] As an embodiment of the present invention, analyzing the browsing behavior preference of the service user based on the user characteristic information includes:

[0086] Determining user identity information corresponding to the service user based on the user characteristic information;

[0087] Based on the user identity information, obtaining cross-platform browsing data corresponding to the service user;

[0088] Extracting multidimensional browsing features from the cross-platform browsing data, performing cluster analysis on the multidimensional browsing features, and obtaining clustered browsing features;

[0089] Performing behavior preference analysis on the clustered browsing features to obtain the browsing behavior preferences of the service users.

[0090] Among them, the user identity information is the personal attribute identification data corresponding to the service user, such as age, occupation, network ID and other basic information used to identify and distinguish individual users; the cross-platform browsing data is a collection of browsing behavior records generated by the service user on multiple network platforms, covering data such as visited pages, browsing time, jump paths, etc.; the multi-dimensional browsing features are key elements extracted from the cross-platform browsing data through multi-dimensional analysis, including content keywords, time patterns, page associations and other features; the clustered browsing features are feature groups with similar behavior patterns formed by aggregating the multi-dimensional browsing features based on data similarity after being processed by a clustering algorithm.

[0091] Furthermore, based on the user feature information, the user identity information corresponding to the service user is determined through identity authentication and feature matching algorithms; based on the user identity information, the cross-platform browsing data corresponding to the service user can be obtained through a cross-platform data authorization retrieval interface; the multidimensional browsing features in the cross-platform browsing data can be extracted through a multimodal feature extraction model, such as extracting visual and semantic features such as images and texts of the browsing page through a convolutional neural network (CNN), analyzing the dynamic features of the browsing time series data using a recurrent neural network (RNN), and then integrating the features of different modalities through a feature fusion layer to obtain multidimensional browsing features including dimensions such as time, content, and interaction; the multidimensional browsing features can be clustered using a density peak clustering algorithm to obtain clustered browsing features; the clustered browsing features can be subjected to behavioral preference analysis to obtain the browsing behavior preferences of the service user, such as using a hierarchical clustering algorithm to perform in-depth analysis of the clustered browsing features to identify the core behavioral patterns therein, and then summarize the browsing behavior preferences of the service user. Specifically, to further intuitively understand the behavioral preference analysis process of the securities information intelligent screening method based on natural language processing in this application, please refer to Figure 2 , which is a schematic diagram of the behavioral preference analysis of the intelligent screening method for securities information based on natural language processing provided by the present invention. It should be noted that, in the present invention, Figure 2 The schematic diagram presented is only used for the analysis and demonstration of the intelligent screening method for securities information based on natural language processing, and is not limited to the analysis and processing of the intelligent screening method for securities information based on natural language processing in different actual application scenarios.

[0092] S2. Combining the browsing behavior preference and the query demand intention, construct the information recommendation tendency of the service user, obtain the securities information database of the securities system, parse the securities information elements corresponding to the securities information database, and calculate the fit between the information recommendation tendency and the securities information elements.

[0093] The present invention constructs the information recommendation tendency of the service user by combining the browsing behavior preference and the query demand intention, which can deeply understand the user's personalized needs and accurately grasp the user's focus and interest in securities information, thereby realizing refined services for users and improving the accuracy and effectiveness of securities information recommendations. Among them, the information recommendation tendency is the direction of the information preference constructed by the service user in combination with the browsing behavior preference and the query demand intention.

[0094] As an embodiment of the present invention, combining the browsing behavior preference and the query demand intention to construct the information recommendation tendency of the service user includes:

[0095] Extracting semantic features of the browsing behavior preference to obtain a behavior semantic feature vector;

[0096] Extracting intent keywords from the query requirement intent and constructing intent feature vectors corresponding to the intent keywords;

[0097] Performing vector fusion processing on the behavior semantic feature vector and the intention feature vector to obtain a comprehensive feature vector;

[0098] Performing feature enhancement processing on the comprehensive feature vector to obtain an enhanced feature vector;

[0099] performing interest mapping processing on the enhanced feature vector to obtain an interest mapping matrix;

[0100] Calculating an interest score corresponding to the service user based on the interest mapping matrix;

[0101] Based on the interest score and the interest mapping matrix, the information recommendation tendency of the service user is constructed.

[0102] Among them, the behavioral semantic feature vector is a set of key semantic information represented in vector form after the browsing behavior preference is extracted through semantic features; the intention keyword is the core word or phrase in the query demand intention that can accurately summarize the user's query target; the intention feature vector is the corresponding intention keyword converted into a numerical vector representation that can be processed by a computer; the comprehensive feature vector is an integrated feature vector after the behavioral semantic feature vector and the intention feature vector are fused through splicing, weighting, etc.; the enhanced feature vector is an enhanced version of the features in the comprehensive feature vector after highlighting key information, suppressing noise, etc.; the interest mapping matrix is ​​the result of presenting the correspondence between features and user interest categories in matrix form after the enhanced feature vector is subjected to interest mapping processing; the interest score is a quantitative indicator corresponding to the service user calculated based on the interest mapping matrix, reflecting the user's interest in different information categories.

[0103] Furthermore, semantic features of the browsing behavior preference can be extracted using a pre-trained language model (such as BERT and RoBERTa) to obtain a behavior semantic feature vector. Intent keywords in the query requirement intent can be extracted using regular expression matching combined with part-of-speech tagging. Intent feature vectors corresponding to the intent keywords can be constructed using word embedding technology (such as Word2Vec and GloVe). The behavior semantic feature vector and the intent feature vector can be fused using feature concatenation combined with an attention mechanism to obtain a comprehensive feature vector. The comprehensive feature vector can be enhanced using autoencoder reconstruction and feature weighting to obtain an enhanced feature vector. The enhanced feature vector can be subjected to interest mapping using a multidimensional association rule mining algorithm to obtain an interest mapping matrix. For example, the Apriori algorithm can be used to mine the association between different features in the enhanced feature vector and the user's interest categories, and the frequent itemsets and their support and confidence can be presented in matrix form to obtain an interest mapping matrix. Based on the interest scores and the interest mapping matrix, the information recommendation tendency of the service user is constructed. Based on the numerical values ​​of the interest scores, high-scoring interests are screened from the interest mapping matrix, and the corresponding information categories are determined, thereby obtaining the information recommendation tendency of the service user.

[0104] Furthermore, as an embodiment of the present invention, the calculating the interest score corresponding to the service user based on the interest mapping matrix includes:

[0105] Calculating the correlation between each matrix element in the interest mapping matrix to obtain the element correlation;

[0106] Normalizing the interest mapping matrix to obtain a normalized interest matrix;

[0107] Performing weight assignment on each matrix element in the normalized interest matrix to obtain element weight;

[0108] Combining the element relevance and the element weight, the interest score corresponding to the service user is calculated using the following formula:

[0109]

[0110] Among them, A represents the interest score of the service user, D a，a+1 represents the element correlation between the ath element and the a+1th element in the normalized interest matrix, E a represents the element weight of the ath element in the normalized interest matrix, λ represents the time decay coefficient, a represents the element sequence number of the normalized interest matrix, and q represents the number of elements in the normalized interest matrix.

[0111] Among them, the element association degree represents the correlation measure between each matrix element in the interest mapping matrix; the normalized interest matrix is ​​the matrix after the interest mapping matrix is ​​processed to map the element values ​​to a specific interval; the element weight is the importance coefficient corresponding to each matrix element in the normalized interest matrix; the time decay coefficient represents the speed control parameter of the interest corresponding to the weakening over time. By analyzing a large amount of user behavior data and observing the law of user interest changes over time, the time decay coefficient can be fitted. Specifically, the user's behavior data on different items at different time points can be collected, such as clicks, purchases, browsing time, etc., and then a mathematical model can be established with time as a variable to try to find a coefficient that can best describe the decay of user interest over time. For example, regression analysis and other methods can be used to fit the user's interest score with time to obtain the optimal time decay coefficient.

[0112] Furthermore, the correlation between each matrix element in the interest mapping matrix can be calculated by the cosine similarity algorithm to obtain the element correlation; the interest mapping matrix can be normalized by the Min-Max normalization method to obtain a normalized interest matrix; and each matrix element in the normalized interest matrix can be weighted by setting a weight rule based on the TF-IDF value or expert experience to obtain the element weight.

[0113] By parsing the securities information elements corresponding to the securities information database, the present invention can obtain structured information such as key features, subject categories, data tags, etc. of securities information, thereby providing data support for the subsequent calculation of the degree of fit between the information recommendation tendency and the securities information elements. The securities information database refers to a collection of various types of information data stored in the securities system, including but not limited to real-time market data, corporate financial report information, industry research reports, policy interpretations, etc. The securities information elements refer to key attributes that constitute the content of securities information, such as information type (research report / news / announcement), securities involved, release time, core views, risk warnings, etc. These elements can reflect the core value and applicable scenarios of securities information. Furthermore, the securities information database can be retrieved from the securities system through the API interface.

[0114] As an embodiment of the present invention, parsing the securities information elements corresponding to the securities information database includes:

[0115] Identifying the securities information tag corresponding to the securities information database and extracting information entity data from the securities information database;

[0116] Performing attribute annotation on the information entity data to obtain information entity attributes;

[0117] Filtering out representative information tags from the securities information tags, and calculating similarity coefficients between the representative information tags and the information entity attributes;

[0118] extracting key information entities from the information entity data based on the similarity coefficient;

[0119] The securities information elements corresponding to the securities information database are parsed from the key information entities.

[0120] Among them, the securities information tag is a mark corresponding to the securities information database for identifying the subject, characteristics, etc. of securities information; the information entity data is data on specific securities-related objects, events, concepts, etc. contained in the securities information database; the information entity attributes are descriptions of the properties, characteristics, parameters, etc. of the information entity data; the representational information tag is a tag in the securities information tag that can highlight the core content and important characteristics of securities information; the similarity coefficient represents a quantitative indicator of the degree of association between the representational information tag and the information entity attribute in terms of semantics, logic, characteristics, etc.; the key information entity is an entity in the information entity data that is highly correlated with the representational information tag and has important significance and value for understanding securities information.

[0121] Furthermore, the securities information tags corresponding to the securities information database can be identified by a tag recognition tool, and the tag recognition tool is compiled by a scripting language; the information entity data in the securities information database can be extracted through a natural language processing (NLP) library (such as NLTK); the information entity data can be attributed by a pre-trained attribute annotation model (such as a rule-based attribute classifier) ​​or combined with a financial field knowledge graph to obtain information entity attributes; the representative information tags in the securities information tags can be screened out by the TextRank algorithm; the similarity coefficient between the representative information tags and the information entity attributes can be calculated by the cosine similarity algorithm; based on the similarity coefficient, key information entities can be extracted from the information entity data by setting a threshold filtering method; the securities information elements corresponding to the securities information database can be parsed from the key information entities, such as company financial indicators, industry policy trends, market risk signals, upstream and downstream relationships in the industrial chain, the impact range of major events, and other structured information that can be used for analysis and decision-making.

[0122] The present invention calculates the degree of fit between the information recommendation tendency and the securities information elements, thereby understanding the degree of match between the service user's personalized interests and the securities information content, and accurately pushes securities information that meets the service user's investment preferences and information needs to the service user, effectively improving the service user's efficiency in obtaining information and the accuracy of investment decisions. The degree of fit represents the degree of match between the information recommendation tendency and the securities information elements. The higher the value, the higher the degree of fit between the information recommendation tendency and the securities information elements, and the more suitable it is for recommendation to users; the lower the value, the lower the degree of match.

[0123] As an embodiment of the present invention, the calculating of the degree of fit between the information recommendation tendency and the securities information elements includes:

[0124] Performing vectorization processing on the information recommendation tendency and the securities information elements respectively to obtain a recommendation tendency vector and an information element vector;

[0125] Combining the recommendation tendency vector and the information element vector, the degree of fit between the information recommendation tendency and the securities information element is calculated using the following formula:

[0126]

[0127] Among them, B represents the degree of fit between information recommendation tendency and securities information elements. represents the recommendation tendency vector, represents the information element vector, The vector modulus of the recommendation tendency vector, Indicates the vector modulus of the information element vector.

[0128] Among them, the recommendation tendency vector and the information element vector are respectively quantitative representations of the information recommendation tendency and the securities information element in the form of vectors in the mathematical space; further, the information recommendation tendency and the securities information element can be vectorized by an encoder to obtain the recommendation tendency vector and the information element vector.

[0129] S3. When the degree of fit is greater than a preset degree of fit, initial securities information is screened out from the securities information database, historical securities trading data of the service user is dispatched in the securities system, securities trading behavior characteristics of the service user are analyzed based on the historical securities trading data, and information filtering processing is performed on the initial securities information based on the securities trading behavior characteristics to obtain filtered securities information.

[0130] The present invention analyzes the securities trading behavior characteristics of the service users based on the historical securities trading data, thereby understanding the investment habits and trading patterns of the service users, and providing an important basis for subsequent information filtering and processing of the initial securities information. Among them, the preset fit is a standard threshold value pre-set according to business needs and user surveys to measure the degree of match between information and user needs; the initial securities information is a set of information that has a certain correlation with user needs, which is preliminarily screened out from the securities information database under the condition that the fit is greater than a preset value; the historical securities trading data is data such as various securities trading records, position information, etc. generated by service users in the past period of time, which are dispatched from the securities system; the securities trading behavior characteristics are behavioral characteristics that can reflect their investment habits, risk preferences, trading strategies, etc., extracted by analyzing the historical securities trading data of service users.

[0131] As an embodiment of the present invention, analyzing the securities trading behavior characteristics of the service user based on the historical securities trading data includes:

[0132] Performing data cleaning on the historical securities trading data to obtain target securities trading data;

[0133] Classifying and integrating the target securities trading data to obtain integrated securities trading data;

[0134] Performing basic feature statistics on the integrated securities trading data to obtain initial features of securities trading, and analyzing trading behaviors corresponding to the initial features of securities trading;

[0135] analyzing feature correlation relationships among the initial features of securities trading, and constructing a feature correlation network among the initial features of securities trading based on the feature correlation relationships;

[0136] Based on the feature association network, core trade features are screened out from the initial features of the securities trade;

[0137] The trading behavior and the core trading characteristics are combined to generate the securities trading behavior characteristics of the service user.

[0138] The target securities trade data is data obtained after data cleaning (e.g., filling missing values, removing outliers, removing duplicate values, etc.) of the historical securities trade data; the integrated securities trade data is structured data obtained after classifying and aggregating the target securities trade data according to dimensions such as trading time, security type, and trading type (e.g., grouping and summarizing); the initial securities trade features are features obtained from the integrated securities trade data by extracting indicators such as trading frequency, holding period, and capital turnover rate and calculating corresponding statistics (e.g., mean, standard deviation, etc.); the trading behavior is the specific trading performance corresponding to the initial securities trade features (e.g., high-frequency trading, long-term holding, etc.); the feature correlation relationship is the degree of correlation (e.g., positive correlation, negative correlation, causal relationship, etc.) obtained by performing causal tests between the initial securities trade features; the feature correlation network is a network structure constructed based on the correlation relationships between the initial securities trade features (e.g., displaying the connections between features through a graph structure); and the core trade features are features that are key to trading behavior and are screened from the initial securities trade features based on the feature correlation network.

[0139] Furthermore, the historical securities trade data can be cleaned by a box plot method to obtain target securities trade data; the target securities trade data can be classified and integrated by a data classification and aggregation algorithm (such as grouping and summarizing by transaction time, securities type, and transaction type) to obtain integrated securities trade data; basic feature statistics of the integrated securities trade data can be performed by a statistical analysis method (such as calculating the mean and standard deviation of indicators such as transaction frequency, holding period, and capital turnover rate) to obtain initial features of securities trade; the trading behavior corresponding to the initial features of securities trade can be analyzed by a pattern recognition algorithm (such as a rule-based matching or machine learning classification model); and the integration of the target securities trade data can be performed by a correlation analysis method (such as a correlation analysis method). The feature association relationship between the initial features of securities trade can be analyzed by using a Wilson correlation coefficient calculation) or a causal test (such as a Granger causality test); based on the feature association relationship, a feature association network between the initial features of securities trade can be constructed by a graph construction algorithm (such as using features as nodes and association relationships as edges to construct a graph structure); based on the feature association network, core trade features can be screened out from the initial features of securities trade by using a graph node importance evaluation algorithm (such as a graph attention mechanism); based on the trade behavior and the core trade features, the securities trade behavior features of the service user can be generated by combining the trade behavior and the core trade features through feature fusion and semantic annotation algorithms (such as associating core features with trade behavior patterns and assigning behavior semantic labels).

[0140] The present invention filters the initial securities information based on the securities trading behavior characteristics to obtain filtered securities information, thereby eliminating redundant information in the initial securities information and improving the value of the information. The filtered securities information is the securities information obtained after the initial securities information is eliminated based on the securities trading behavior characteristics. Furthermore, based on the securities trading behavior characteristics, the initial securities information is filtered to obtain filtered securities information, such as filtering information that is not timely enough based on user trading frequency, eliminating irrelevant industry information based on investment preferences, and screening out high-risk or low-return content based on risk tolerance.

[0141] S4. Query the user financial information corresponding to the service user, evaluate the risk tolerance corresponding to the service user based on the user financial information, and optimize the filtered securities information based on the risk tolerance to obtain the filtered securities information of the service user.

[0142] The present invention evaluates the risk tolerance corresponding to the service user based on the user's financial information. The risk tolerance can be used to understand the risk tolerance of the service user, thereby preventing the user from taking excessive risks due to receiving inappropriate securities information, thereby improving the effectiveness and accuracy of the information. Among them, the user financial information refers to the comprehensive data of various assets, liabilities and income of the service user, including but not limited to bank deposits, real estate, financial investment product holdings, personal debts, average monthly income, etc. The risk tolerance refers to the service user's ability to withstand the risk losses that may be faced in investment activities. Furthermore, the user financial information corresponding to the service user can be queried through the user-authorized data interface to connect to the banking, securities, tax and other financial institution systems.

[0143] As an embodiment of the present invention, the step of evaluating the risk tolerance of the service user based on the user's financial information includes:

[0144] Verifying the user's asset information to obtain trusted asset information;

[0145] Classify the trusted asset information to obtain liability asset information and income asset information;

[0146] Extracting the debt scale and debt structure from the debt asset information, and calculating the debt stress index corresponding to the service user based on the debt scale and the debt structure;

[0147] Extracting the income scale and income composition from the income asset information, and calculating the income elasticity index corresponding to the service user based on the income scale and the income composition;

[0148] The risk tolerance of the service user is evaluated by combining the debt pressure index and the income elasticity index.

[0149] Among them, the trusted asset information is the data that is verified and processed to ensure the authenticity and reliability of the user asset information; the liability asset information and the income asset information are two types of data sets classified according to nature in the trusted asset information; the debt scale and the debt structure are key characteristic parameters in the liability asset information; the debt pressure index indicates the degree of debt repayment pressure corresponding to the service user; the income scale and the income structure are the core analysis elements in the income asset information; the income elasticity index indicates the ability of the service user's corresponding income to cope with risk fluctuations.

[0150] Furthermore, the user's asset information can be verified and processed through blockchain evidence storage, cross-data verification and digital signature technology to obtain trusted asset information; the trusted asset information can be classified and processed according to the nature of the assets and the flow of funds through machine learning classification algorithms to obtain liability asset information and income asset information; the debt scale and debt structure in the liability asset information can be extracted through statistical analysis tools, such as Tableau tools; the debt pressure index corresponding to the service user is calculated based on the debt scale and the debt structure, and the calculation steps are as follows: first, the balance of various types of user debts is added to obtain the total debt, which is used to measure the debt scale; then, the proportion of short-term debt and long-term debt in the debt structure is analyzed. The higher the proportion of short-term debt, the greater the liquidity pressure; finally, the debt pressure index is comprehensively assessed based on the size of the total debt and the proportion of short-term debt. The higher the total debt and the greater the proportion of short-term debt, the higher the debt pressure index; it can be used The above-mentioned Tableau tool extracts the income scale and income structure from the income asset information; combined with the income scale and the income structure, the income elasticity index corresponding to the service user is calculated, and the calculation steps are as follows: first, summarize the various types of user income to obtain the total income scale to clarify the size of the income; then analyze the proportion of different income types (such as wages, investment income, rent, etc.) in the income structure. The higher the proportion of income types with large fluctuations, the greater the potential elasticity; finally, the comprehensive income scale and the proportion of high-volatility income are considered. The smaller the income scale and the higher the proportion of high-volatility income, the higher the income elasticity index, and vice versa; combined with the debt pressure index and the income elasticity index, the risk tolerance corresponding to the service user is evaluated. If the debt pressure index is low and the income elasticity index is high, it means that the user has low debt repayment pressure, great income growth potential, and high risk tolerance; conversely, if the debt pressure index is high and the income elasticity index is low, it means that the user has weak risk resistance and low risk tolerance.

[0151] The present invention optimizes the filtered securities information based on the risk tolerance to obtain the filtered securities information for the service users, thereby improving the effectiveness of the securities information. Furthermore, based on the risk tolerance, the filtered securities information is optimized to obtain the filtered securities information for the service users. For example, for users with high risk tolerance, the focus is on pushing in-depth analysis information on securities with high growth potential but large volatility, and for users with low risk tolerance, the focus is on information interpretation of low-volatility, stable products, thereby achieving accurate matching of securities information and user risk preferences.

[0152] Compared to the problems described in the background art, the present invention, by parsing the query intent and user characteristic information corresponding to the service user based on the request instruction, can convert the diverse instructions of the service user into structured information that the system can understand, thereby avoiding subsequent service deviations. Furthermore, the present invention, by combining the browsing behavior preferences and the query intent, constructs the information recommendation tendency of the service user, can deeply understand the user's personalized needs, accurately grasp the user's focus and interest in securities information, thereby achieving refined services for users and improving the accuracy and effectiveness of securities information recommendations. The present invention, by analyzing the securities trading behavior characteristics of the service user based on the historical securities trading data, can understand the investment habits and trading patterns of the service user, thereby providing an important basis for subsequent information filtering of the initial securities information. Furthermore, the present invention, by evaluating the risk tolerance corresponding to the service user based on the user's financial information, can understand the risk tolerance of the service user through the risk tolerance, thereby preventing the user from taking excessive risks due to receiving inappropriate securities information, thereby improving the effectiveness and accuracy of the information. Therefore, the intelligent securities information screening method and system based on natural language processing provided by the embodiments of the present invention can improve the accuracy of securities information screening.

[0153] Example 2:

[0154] like Figure 3 1 is a functional module diagram of a securities information intelligent screening system based on natural language processing according to the present invention.

[0155] The natural language processing-based intelligent securities information screening system 300 described in the present invention can be installed in an electronic device. Depending on the functionality implemented, the natural language processing-based intelligent securities information screening system can include a behavior preference analysis module 301, a compatibility calculation module 302, an information filtering module 303, and an information optimization module 304. A module, also referred to as a unit, is a series of computer program segments that can be executed by an electronic device processor and perform a fixed function, and is stored in the electronic device's memory.

[0156] In the embodiment of the present invention, the functions of each module / unit are as follows:

[0157] The behavior preference analysis module 301 is configured to receive a request instruction from a service user through the securities system, parse the request instruction to determine the query demand intention and user characteristic information corresponding to the service user, and analyze the browsing behavior preference of the service user based on the user characteristic information;

[0158] The compatibility calculation module 302 is configured to combine the browsing behavior preference and the query demand intention to construct the information recommendation tendency of the service user, obtain the securities information database of the securities system, parse the securities information elements corresponding to the securities information database, and calculate the compatibility between the information recommendation tendency and the securities information elements;

[0159] The information filtering module 303 is configured to, when the degree of compatibility is greater than a preset degree of compatibility, filter initial securities information from the securities information database, dispatch historical securities trading data of the service user in the securities system, analyze the securities trading behavior characteristics of the service user based on the historical securities trading data, and perform information filtering on the initial securities information based on the securities trading behavior characteristics to obtain filtered securities information;

[0160] The information optimization module 304 is used to query the user asset information corresponding to the service user, evaluate the risk tolerance corresponding to the service user based on the user asset information, and optimize the filtered securities information based on the risk tolerance to obtain the filtered securities information of the service user.

[0161] In detail, each module in the securities information intelligent screening system 200 based on natural language processing in the embodiment of the present invention adopts the same method as above when in use. Figure 1 The technical means are the same as the intelligent screening method of securities information based on natural language processing described in, and can produce the same technical effects, so I will not go into details here.

[0162] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent screening of securities information based on natural language processing, characterized in that: The method comprises: Utilizing the securities system to receive a request instruction from a service user, parsing the query demand intention and user characteristic information corresponding to the service user based on the request instruction, and analyzing the browsing behavior preference of the service user based on the user characteristic information; Combining the browsing behavior preference and the query demand intention, constructing the information recommendation tendency of the service user, obtaining the securities information database of the securities system, parsing the securities information elements corresponding to the securities information database, and calculating the degree of fit between the information recommendation tendency and the securities information elements; When the degree of compatibility is greater than a preset degree of compatibility, initial securities information is screened from the securities information database, historical securities trading data of the service user is dispatched in the securities system, securities trading behavior characteristics of the service user are analyzed based on the historical securities trading data, and information filtering processing is performed on the initial securities information based on the securities trading behavior characteristics to obtain filtered securities information; The user asset information corresponding to the service user is queried, and based on the user asset information, the risk tolerance corresponding to the service user is evaluated; based on the risk tolerance, the filtered securities information is optimized to obtain the filtered securities information of the service user.

2. The method for intelligent screening of securities information based on natural language processing according to claim 1, characterized in that: The step of parsing the query demand intention and user characteristic information corresponding to the service user based on the request instruction includes: Performing protocol parsing on the request instruction to obtain a request instruction packet; Performing word sense disambiguation on the request instruction packet to obtain a request instruction text; Performing intent classification on the request instruction text to obtain a classified instruction text; Extracting parameters from the classified instruction text to obtain instruction text parameters; Based on the instruction text parameters, the query demand intention and user characteristic information corresponding to the service user are analyzed.

3. The method for intelligent screening of securities information based on natural language processing according to claim 1, characterized in that: Analyzing the browsing behavior preferences of the service user based on the user characteristic information includes: Determining user identity information corresponding to the service user based on the user characteristic information; Based on the user identity information, obtaining cross-platform browsing data corresponding to the service user; Extracting multidimensional browsing features from the cross-platform browsing data, performing cluster analysis on the multidimensional browsing features, and obtaining clustered browsing features; Performing behavior preference analysis on the clustered browsing features to obtain the browsing behavior preferences of the service users.

4. The method for intelligent screening of securities information based on natural language processing according to claim 1, characterized in that: Combining the browsing behavior preference and the query demand intention to construct the information recommendation tendency of the service user includes: Extracting semantic features of the browsing behavior preference to obtain a behavior semantic feature vector; Extracting intent keywords from the query requirement intent and constructing intent feature vectors corresponding to the intent keywords; Performing vector fusion processing on the behavior semantic feature vector and the intention feature vector to obtain a comprehensive feature vector; Performing feature enhancement processing on the comprehensive feature vector to obtain an enhanced feature vector; performing interest mapping processing on the enhanced feature vector to obtain an interest mapping matrix; Calculating an interest score corresponding to the service user based on the interest mapping matrix; Based on the interest score and the interest mapping matrix, the information recommendation tendency of the service user is constructed.

5. The method for intelligent screening of securities information based on natural language processing according to claim 4, characterized in that: The calculating, based on the interest mapping matrix, the interest score corresponding to the service user includes: Calculating the correlation between each matrix element in the interest mapping matrix to obtain the element correlation; Normalizing the interest mapping matrix to obtain a normalized interest matrix; Performing weight assignment on each matrix element in the normalized interest matrix to obtain element weight; Combining the element relevance and the element weight, the interest score corresponding to the service user is calculated using the following formula: Among them, A represents the interest score of the service user, D a，a+1 represents the element correlation between the ath element and the a+1th element in the normalized interest matrix, E a represents the element weight of the ath element in the normalized interest matrix, λ represents the time decay coefficient, a represents the element sequence number of the normalized interest matrix, and q represents the number of elements in the normalized interest matrix.

6. The method for intelligent screening of securities information based on natural language processing according to claim 1, characterized in that: The analyzing of securities information elements corresponding to the securities information database includes: Identifying the securities information tag corresponding to the securities information database and extracting information entity data from the securities information database; Performing attribute annotation on the information entity data to obtain information entity attributes; Filtering out representative information tags from the securities information tags, and calculating similarity coefficients between the representative information tags and the information entity attributes; extracting key information entities from the information entity data based on the similarity coefficient; The securities information elements corresponding to the securities information database are parsed from the key information entities.

7. The method for intelligent screening of securities information based on natural language processing according to claim 1, characterized in that: The calculating of the degree of fit between the information recommendation tendency and the securities information elements includes: Performing vectorization processing on the information recommendation tendency and the securities information elements respectively to obtain a recommendation tendency vector and an information element vector; Combining the recommendation tendency vector and the information element vector, the degree of fit between the information recommendation tendency and the securities information element is calculated using the following formula: Among them, B represents the degree of fit between information recommendation tendency and securities information elements. represents the recommendation tendency vector, represents the information element vector, The vector modulus of the recommendation tendency vector, Indicates the vector modulus of the information element vector.

8. The method for intelligent screening of securities information based on natural language processing according to claim 1, characterized in that: The analyzing the securities trading behavior characteristics of the service user based on the historical securities trading data includes: Performing data cleaning on the historical securities trading data to obtain target securities trading data; Classifying and integrating the target securities trading data to obtain integrated securities trading data; Performing basic feature statistics on the integrated securities trading data to obtain initial features of securities trading, and analyzing trading behaviors corresponding to the initial features of securities trading; analyzing feature correlation relationships among the initial features of securities trading, and constructing a feature correlation network among the initial features of securities trading based on the feature correlation relationships; Based on the feature association network, core trade features are screened out from the initial features of the securities trade; The trading behavior and the core trading characteristics are combined to generate the securities trading behavior characteristics of the service user.

9. The method for intelligent screening of securities information based on natural language processing according to claim 1, characterized in that: The step of evaluating the risk tolerance of the service user based on the user's financial information includes: Verifying the user's asset information to obtain trusted asset information; Classify the trusted asset information to obtain liability asset information and income asset information; Extracting the debt scale and debt structure from the debt asset information, and calculating the debt stress index corresponding to the service user based on the debt scale and the debt structure; Extracting the income scale and income composition from the income asset information, and calculating the income elasticity index corresponding to the service user based on the income scale and the income composition; The risk tolerance of the service user is evaluated by combining the debt pressure index and the income elasticity index.

10. A securities information intelligent screening system based on natural language processing, characterized in that: The system comprises: a behavior preference analysis module configured to receive a request instruction from a service user using the securities system, parse the request instruction to determine the query demand intention and user characteristic information corresponding to the service user, and analyze the browsing behavior preference of the service user based on the user characteristic information; a compatibility calculation module, configured to combine the browsing behavior preference and the query demand intention to construct the information recommendation tendency of the service user, obtain the securities information database of the securities system, parse the securities information elements corresponding to the securities information database, and calculate the compatibility between the information recommendation tendency and the securities information elements; an information filtering module, configured to, when the degree of compatibility is greater than a preset degree of compatibility, filter initial securities information from the securities information database, dispatch historical securities trading data of the service user in the securities system, analyze securities trading behavior characteristics of the service user based on the historical securities trading data, and perform information filtering on the initial securities information based on the securities trading behavior characteristics to obtain filtered securities information; The information optimization module is used to query the user asset information corresponding to the service user, evaluate the risk tolerance corresponding to the service user based on the user asset information, and optimize the filtered securities information based on the risk tolerance to obtain the filtered securities information of the service user.

Images