A data analysis-based risk control message intelligent matching system and method

Abstract

The application discloses a kind of based on data analysis's risk control message intelligent matching system and method, it is related to message matching technical field, the matching method includes the following steps: acquisition comes from the risk control message and behavior data of multiple source channels, constructs several risk control data sets;Any risk control data set is identified to the risk control scene where it is located;Risk control message in any risk control data set is carried out semantic analysis, extracts several risk control features, and outputs the intention label corresponding to risk control data set;All risk control messages are carried out homotype clustering;The feature weight of each risk control feature under different risk control scenes is adjusted;Real-time acquisition risk control message is carried out feature extraction and scene identification, and each feature is weighted operation to generate real-time risk value;Based on the preset risk assessment threshold, abnormal risk control message is early warned and pushed;The coverage of risk detection is greatly improved with judgment accuracy, and the intelligent and precision level of risk control system is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of message matching technology, specifically a risk control message intelligent matching system and method based on data analysis. Background Technology

[0002] With the rapid development of various digital platforms such as finance, e-commerce, and social media, massive amounts of data, such as user behavior, transaction records, and interactive messages, contain various risk signals. Traditional risk control systems rely heavily on rule engines and manual review, but in scenarios with diverse message content and varied semantics, it is difficult to accurately identify risk intentions, which can easily lead to the underreporting of important risk events or the misinterpretation of normal behavior.

[0003] Existing technologies attempt to classify risk events using clustering and classification algorithms. However, these methods are usually unable to effectively distinguish the differences in risks under different scenarios, nor can they dynamically assess the weight of different risk characteristics. They also lack the ability to perceive and adaptively adjust to the real-time risk situation, resulting in low risk response efficiency and limited overall system intelligence. Summary of the Invention

[0004] The purpose of this invention is to provide a data analysis-based intelligent matching system and method for risk control messages, in order to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a data analysis-based intelligent matching method for risk control messages, the matching method comprising the following steps:

[0006] Step S100: Collect risk control messages and behavioral data from multiple sources to construct several risk control datasets; analyze the behavioral data in the risk control datasets and identify the risk control scenario in which any risk control dataset is located;

[0007] Step S200: Perform semantic parsing on the risk control messages in any risk control dataset to extract several risk control features; perform combined analysis on the several risk control features to output the intent label corresponding to the risk control dataset;

[0008] Step S300: Based on risk control scenarios and intent tags, perform homogeneous clustering on all risk control messages; compare homogeneous risk control messages and adjust the feature weights of each risk control feature under different risk control scenarios;

[0009] Step S400: Extract features and identify scenarios from the real-time collected risk control messages; perform weighted calculations on each feature based on the risk control scenario to generate a real-time risk value; and push early warnings to abnormal risk control messages based on preset risk assessment thresholds.

[0010] Furthermore, step S100 includes the following steps:

[0011] Step S101: Pre-set several information collection channels to collect risk control messages and user behavior data. At each unit period, summarize the risk control messages and user behavior data collected from all information collection channels, and clean, format, and normalize the summarized risk control messages and user behavior data to generate a corresponding risk control dataset. The pre-set information collection channels may include transaction platforms, customer service systems, behavior logs, external blacklists, etc. On the transaction platform, user transfer information and payment information can be collected; the customer service system can collect user complaints and inquiry texts; the behavior log can collect user login IP, device fingerprint, and operation sequence; and the external blacklist can collect known risky accounts and high-risk IP databases.

[0012] Step S102: Randomly select a risk control dataset, extract user behavior data from the selected risk control dataset, extract corresponding behavioral features from the arbitrary user behavior data, and obtain several behavioral features of the selected risk control dataset; pre-establish a scenario database, which stores several risk control scenarios, in which each risk control scenario is matched with a corresponding scenario feature set; arbitrarily select a risk control scenario, compare the scenario feature set of the selected risk control scenario with several behavioral features of the selected risk control dataset, if the scenario feature set contains several behavioral features, then set the selected risk control scenario as the risk control scenario in which the selected risk control dataset is located; risk control scenarios include account theft login scenario, fraudulent transaction scenario, money laundering behavior scenario, and spam registration scenario; among them, the behavioral features corresponding to the account theft login scenario include multiple failed logins, login from different locations, and large transactions; the behavioral features corresponding to the fraudulent transaction scenario include high-frequency small-amount transactions, concentrated transaction time, risky list accounts, and immediate refund after transaction.

[0013] Furthermore, step S200 includes the following steps:

[0014] Step S201: Randomly select a risk control dataset, obtain the risk control messages in the selected risk control dataset, and randomly extract one risk control message from it. Use natural language processing technology to perform text content recognition and word segmentation on the extracted risk control message to obtain several keywords.

[0015] Step S202: A feature database is pre-built, storing several risk control features. A keyword is randomly selected from several keywords, and its semantic similarity is compared with each risk control feature. A similarity threshold is preset. If the similarity of several risk control features exceeds the threshold, the risk control feature with the highest similarity is set as the risk control feature corresponding to the selected keyword. The risk control features corresponding to each keyword are obtained, generating a risk control feature set. Risk control features include unauthorized logins, large-amount transfers, identity information leakage, cross-border transactions, etc.

[0016] Step S203: Pre-build an intent tag database and store several intent tags. Any intent tag matches a risk control feature group. Randomly select an intent tag and compare the risk control feature set with the risk control feature group of the selected intent tag. If any risk control feature in the risk control feature set belongs to the risk control feature group, then set the selected intent tag as the target intent tag.

[0017] Step S204: If there are several target intent tags in the risk control feature set, then arbitrarily select one target intent tag, count the number of risk control features in the selected target intent tag as M, obtain the number of risk control features in the risk control feature set as a, and calculate the correlation degree of the selected target intent tag G=a / M; select the target intent tag with the highest correlation degree and set it as the intent tag for extracting risk control messages.

[0018] Furthermore, step S300 includes the following steps:

[0019] Step S301: Obtain the risk control scenario and intent tag of each risk control message. Randomly select two risk control messages. If the risk control scenario and intent tag of the two risk control messages are the same, divide the two risk control messages into the same type of risk control messages. Divide all risk control messages into several sets of the same type of risk control messages.

[0020] Step S302: Arbitrarily select a risk control scenario and an intent label, and extract the set of risk control messages of the same type corresponding to the selected risk control scenario and intent label; set all the risk control messages in the extracted set of risk control messages of the same type as target risk control messages, arbitrarily select a target risk control message, extract several risk control features from the selected target risk control message, arbitrarily select one risk control feature from the several risk control features, and count the number of times the selected risk control feature appears in each risk control message as H, where the number of times the selected risk control feature appears in each target risk control message is h1, and calculate the occurrence frequency f of the selected risk control feature under the selected risk control scenario and intent label as f=h1 / H;

[0021] Step S303: Summarize several risk control features of each target risk control information in the extracted risk control message set to obtain a risk control feature set for selecting risk control scenarios and intent tags. Obtain the occurrence frequency of each risk control feature in the risk control feature set, and set the occurrence frequency of the i-th risk control feature as f. i According to the formula:

[0022] ;

[0023] Where e is the number of risk control features in the risk control feature set; the first weight Q1 of the selected risk control features is calculated; the first weight is used to measure the importance of a certain risk control feature under a specific risk control scenario and intent label;

[0024] Step S304: Extract intent tags containing risk control features in the selected risk control scenario and set them as target intent tags. Count the number of target intent tags and obtain the occurrence frequency of risk control features in any target intent tag. Set the occurrence frequency of risk control features in the j1th type of target intent tag as f(j1), according to the formula:

[0025] ;

[0026] Where g is the number of target intent tags; the second weight Q2 of the selected risk control feature is calculated; the second weight is used to measure the distinguishability or discriminative power of the feature among multiple intent tags;

[0027] Step S305: Obtain the first weight Q1 and the second weight Q2 of the selected risk control feature in the selected risk control scenario, and calculate the feature weight of the selected risk control feature in the selected risk control scenario as Q=Q1×(1+Q2); the feature weight is used for weighted calculation in real-time risk assessment, which can improve the accuracy of risk judgment and scenario adaptability.

[0028] Furthermore, step S400 includes the following steps:

[0029] Step S401: Whenever a risk control message is collected in real time, semantic parsing and word segmentation are performed on the risk control message to extract several real-time keywords; the semantic similarity of each real-time keyword is compared with the risk control features stored in the feature database, and real-time risk control features with similarity exceeding a preset similarity threshold are selected to form a set of real-time risk control features for the risk control message; the set of real-time risk control features is compared with each intent tag in the intent tag database to obtain the real-time intent tag corresponding to the risk control message.

[0030] Step S402: Collect user behavior data when risk control messages are collected in real time, extract several behavioral features from the user behavior data, match the several behavioral features with the set of scene features of various risk control scenarios stored in the scene database, and identify the real-time risk control scenario in which the risk control message is located.

[0031] Step S403: Randomly select the k-th real-time risk control feature from the real-time risk control feature set, and obtain the feature weight Q of the k-th real-time risk control feature under the real-time risk control scenario and real-time intent label. k According to the formula:

[0032] ;

[0033] Where s is the number of real-time risk control features; the real-time risk value R of the real-time collected risk control messages is calculated; and a risk assessment threshold R is preset. th If R≥R th If so, the risk control message will be marked as abnormal and a risk warning will be pushed.

[0034] To better implement the above methods, a risk control message intelligent matching system is also proposed. The matching system includes a risk control data analysis module, a risk intent analysis module, a risk weight division module, and a risk control early warning identification module.

[0035] The risk control data analysis module is used to collect risk control messages and behavioral data from multiple sources to construct several risk control datasets; analyze the behavioral data in the risk control datasets; and identify the risk control scenarios in which any risk control dataset is located.

[0036] The risk intent analysis module is used to perform semantic parsing on risk control messages in any risk control dataset, extract several risk control features, perform combined analysis on several risk control features, and output the intent label corresponding to the risk control dataset.

[0037] The risk weighting module is used to perform homogeneous clustering of all risk control messages based on risk control scenarios and intent tags; compare homogeneous risk control messages, and adjust the feature weights of each risk control feature under different risk control scenarios;

[0038] The risk control early warning and identification module is used to extract features and identify scenarios from real-time collected risk control messages. Based on the risk control scenario, it performs weighted calculations on each feature to generate a real-time risk value. Based on the preset risk assessment threshold, it pushes early warnings for abnormal risk control messages.

[0039] Furthermore, the risk control data analysis module includes a risk control data acquisition unit and a risk control scenario identification unit;

[0040] The risk control data acquisition unit is used to collect risk control messages and behavioral data from multiple sources to construct several risk control datasets; the risk control scenario identification unit is used to analyze the behavioral data in the risk control datasets and identify the risk control scenario in which any risk control dataset is located.

[0041] Furthermore, the risk intent analysis module includes a risk control message parsing unit and a risk intent identification unit;

[0042] The risk control message parsing unit is used to perform semantic parsing on risk control messages in any risk control dataset and extract several risk control features; the risk intent recognition unit is used to perform combined analysis on several risk control features and output the intent label corresponding to the risk control dataset.

[0043] Furthermore, the risk weighting module includes a risk control message clustering unit and a feature weight adjustment unit;

[0044] The risk control message clustering unit is used to perform homogeneous clustering of all risk control messages based on risk control scenarios and intent tags; the feature weight adjustment unit is used to compare homogeneous risk control messages and adjust the feature weights of each risk control feature under different risk control scenarios.

[0045] Furthermore, the risk control early warning identification module includes a feature weighted analysis unit and an abnormal risk control early warning unit;

[0046] The feature weighted analysis unit is used to extract features and identify scenarios from real-time collected risk control messages, and to generate real-time risk values ​​by weighting each feature based on the risk control scenario; the abnormal risk control early warning unit is used to push early warnings for abnormal risk control messages based on preset risk assessment thresholds.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] 1. This invention achieves refined identification and classification of complex risk scenarios by integrating multi-source behavioral data and message semantic parsing, significantly improving the coverage and accuracy of risk detection, and significantly enhancing the intelligence and precision of the risk control system;

[0049] 2. By constructing an association analysis system between intent tags and risk control features, this invention achieves a deep semantic understanding and structured representation of risk intent, which helps the system to more accurately identify the potential risk motivations behind user behavior and provides a more reliable basis for risk warning and decision support.

[0050] 3. This invention performs feature weight analysis on risk control features and automatically optimizes feature importance based on different risk control scenarios and intent tags, making risk assessment more scenario-sensitive and timely; at the same time, it supports real-time data stream processing and online risk assessment, significantly improving the initiative and timeliness of risk prevention and control. Attached Figure Description

[0051] Figure 1 This is a schematic diagram illustrating the steps of a data analysis-based intelligent matching method for risk control messages.

[0052] Figure 2 This is a schematic diagram of a data analysis-based intelligent matching system for risk control messages. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Example: Figures 1 to 2 As shown, this invention provides a data analysis-based intelligent matching method for risk control messages. The matching method includes the following steps:

[0055] Step S100: Collect risk control messages and behavioral data from multiple sources to construct several risk control datasets; analyze the behavioral data in the risk control datasets and identify the risk control scenario in which any risk control dataset is located;

[0056] Step S100 includes the following steps:

[0057] Step S101: Pre-set several information collection channels to collect risk control messages and user behavior data. At each unit period, summarize the risk control messages and user behavior data collected from all information collection channels, and clean, format and normalize the summarized risk control messages and user behavior data to generate a corresponding risk control dataset.

[0058] Step S102: Randomly select a risk control dataset, extract user behavior data from the selected risk control dataset, extract corresponding behavioral features from the arbitrary user behavior data, and obtain several behavioral features of the selected risk control dataset; pre-establish a scenario database, which stores several risk control scenarios, in which each risk control scenario is matched with a corresponding scenario feature set; arbitrarily select a risk control scenario, compare the scenario feature set of the selected risk control scenario with several behavioral features of the selected risk control dataset, and if the scenario feature set contains several behavioral features, then set the selected risk control scenario as the risk control scenario in which the selected risk control dataset is located;

[0059] Example 1: The system extracts behavioral features from a dataset, such as "multiple failed login attempts in a short period of time", "login from a different location", and "large transaction amount". The scenario database pre-stores risk control scenarios and corresponding feature sets: the "account theft login" scenario corresponds to the features of "multiple failed login attempts", "large transaction amount" and "login from a different location". Based on feature comparison, the risk control scenario of a dataset is the account theft login scenario.

[0060] Step S200: Perform semantic parsing on the risk control messages in any risk control dataset to extract several risk control features; perform combined analysis on the several risk control features to output the intent label corresponding to the risk control dataset;

[0061] Step S200 includes the following steps:

[0062] Step S201: Randomly select a risk control dataset, obtain the risk control messages in the selected risk control dataset, and randomly extract one risk control message from it. Use natural language processing technology to perform text content recognition and word segmentation on the extracted risk control message to obtain several keywords.

[0063] Step S202: Pre-build a feature database to store several risk control features. Randomly select a keyword from several keywords and compare the similarity between the selected keyword and each risk control feature in terms of word meaning. Obtain the similarity between the selected keyword and each risk control feature. Preset a similarity threshold. If the similarity of several risk control features exceeds the similarity threshold, set the risk control feature with the highest similarity as the risk control feature corresponding to the selected keyword. Obtain the risk control features corresponding to each keyword and generate a risk control feature set.

[0064] Step S203: Pre-build an intent tag database and store several intent tags. Any intent tag matches a risk control feature group. Randomly select an intent tag and compare the risk control feature set with the risk control feature group of the selected intent tag. If any risk control feature in the risk control feature set belongs to the risk control feature group, then set the selected intent tag as the target intent tag.

[0065] Step S204: If there are several target intent tags in the risk control feature set, then arbitrarily select one target intent tag, count the number of risk control features in the selected target intent tag as M, obtain the number of risk control features in the risk control feature set as a, and calculate the correlation degree of the selected target intent tag G=a / M; select the target intent tag with the highest correlation degree and set it as the intent tag for extracting risk control messages.

[0066] Example 2: A user sends a message saying "My account has been hacked, and 5,000 yuan was spent overseas." A similarity comparison with risk control features in the feature database shows that "hacking" matches the risk control feature "account anomaly," "overseas" matches the risk control feature "transactions in other locations," and "spending 5,000 yuan" matches the risk control feature "large transaction." The final risk control feature set is: account anomaly, transactions in other locations, and large transaction. In the intent tag database, the "fraudulent spending intent" tag corresponds to the feature group of account anomaly, transactions in other locations, and large transaction. Its correlation degree is calculated as G=3 / 3=1, higher than other tags. Therefore, this message is labeled with the "fraudulent spending intent" tag.

[0067] Step S300: Based on risk control scenarios and intent tags, perform homogeneous clustering on all risk control messages; compare homogeneous risk control messages and adjust the feature weights of each risk control feature under different risk control scenarios;

[0068] Step S300 includes the following steps:

[0069] Step S301: Obtain the risk control scenario and intent tag of each risk control message. Randomly select two risk control messages. If the risk control scenario and intent tag of the two risk control messages are the same, divide the two risk control messages into the same type of risk control messages. Divide all risk control messages into several sets of the same type of risk control messages.

[0070] Step S302: Arbitrarily select a risk control scenario and an intent label, and extract the set of risk control messages of the same type corresponding to the selected risk control scenario and intent label; set all the risk control messages in the extracted set of risk control messages of the same type as target risk control messages, arbitrarily select a target risk control message, extract several risk control features from the selected target risk control message, arbitrarily select one risk control feature from the several risk control features, and count the number of times the selected risk control feature appears in each risk control message as H, where the number of times the selected risk control feature appears in each target risk control message is h1, and calculate the occurrence frequency f of the selected risk control feature under the selected risk control scenario and intent label as f=h1 / H;

[0071] Step S303: Summarize several risk control features of each target risk control information in the extracted risk control message set to obtain a risk control feature set for selecting risk control scenarios and intent tags. Obtain the occurrence frequency of each risk control feature in the risk control feature set, and set the occurrence frequency of the i-th risk control feature as f. i According to the formula:

[0072] ;

[0073] Where e is the number of risk control features in the risk control feature set; the first weight Q1 for selecting risk control features is calculated;

[0074] Step S304: Extract intent tags containing risk control features in the selected risk control scenario and set them as target intent tags. Count the number of target intent tags and obtain the occurrence frequency of risk control features in any target intent tag. Set the occurrence frequency of risk control features in the j1th type of target intent tag as f(j1), according to the formula:

[0075] ;

[0076] Where g is the number of target intent tags; the second weight Q2 for selecting risk control features is calculated;

[0077] Step S305: Obtain the first weight Q1 and the second weight Q2 of the selected risk control feature in the selected risk control scenario, and calculate the feature weight of the selected risk control feature in the selected risk control scenario as Q=Q1×(1+Q2).

[0078] Step S400: Extract features and identify scenarios from the real-time collected risk control messages; perform weighted calculations on each feature based on the risk control scenario to generate a real-time risk value; and push early warnings to abnormal risk control messages based on preset risk assessment thresholds.

[0079] Step S400 includes the following steps:

[0080] Step S401: Whenever a risk control message is collected in real time, semantic parsing and word segmentation are performed on the risk control message to extract several real-time keywords; the semantic similarity of each real-time keyword is compared with the risk control features stored in the feature database, and real-time risk control features with similarity exceeding a preset similarity threshold are selected to form a set of real-time risk control features for the risk control message; the set of real-time risk control features is compared with each intent tag in the intent tag database to obtain the real-time intent tag corresponding to the risk control message.

[0081] Step S402: Collect user behavior data when risk control messages are collected in real time, extract several behavioral features from the user behavior data, match the several behavioral features with the set of scene features of various risk control scenarios stored in the scene database, and identify the real-time risk control scenario in which the risk control message is located.

[0082] Step S403: Randomly select the k-th real-time risk control feature from the real-time risk control feature set, and obtain the feature weight Q of the k-th real-time risk control feature under the real-time risk control scenario and real-time intent label. k According to the formula:

[0083] ;

[0084] Where s is the number of real-time risk control features; the real-time risk value R of the real-time collected risk control messages is calculated; and a risk assessment threshold R is preset. th If R≥R th If so, the risk control message will be marked as abnormal and a risk warning will be pushed.

[0085] A risk control message intelligent matching system, the matching system includes a risk control data analysis module, a risk intent analysis module, a risk weighting module, and a risk control early warning identification module;

[0086] The risk control data analysis module is used to collect risk control messages and behavioral data from multiple sources to construct several risk control datasets; analyze the behavioral data in the risk control datasets; and identify the risk control scenarios in which any risk control dataset is located.

[0087] The risk intent analysis module is used to perform semantic parsing on risk control messages in any risk control dataset, extract several risk control features, perform combined analysis on several risk control features, and output the intent label corresponding to the risk control dataset.

[0088] The risk weighting module is used to perform homogeneous clustering of all risk control messages based on risk control scenarios and intent tags; compare homogeneous risk control messages, and adjust the feature weights of each risk control feature under different risk control scenarios;

[0089] The risk control early warning and identification module is used to extract features and identify scenarios from real-time collected risk control messages. Based on the risk control scenario, it performs weighted calculations on each feature to generate a real-time risk value. Based on the preset risk assessment threshold, it pushes early warnings for abnormal risk control messages.

[0090] The risk control data analysis module includes a risk control data acquisition unit and a risk control scenario identification unit.

[0091] The risk control data acquisition unit is used to collect risk control messages and behavioral data from multiple sources to construct several risk control datasets; the risk control scenario identification unit is used to analyze the behavioral data in the risk control datasets and identify the risk control scenario in which any risk control dataset is located.

[0092] The risk intent analysis module includes a risk control message parsing unit and a risk intent identification unit.

[0093] The risk control message parsing unit is used to perform semantic parsing on risk control messages in any risk control dataset and extract several risk control features; the risk intent recognition unit is used to perform combined analysis on several risk control features and output the intent label corresponding to the risk control dataset.

[0094] The risk weighting module includes a risk control message clustering unit and a feature weight adjustment unit.

[0095] The risk control message clustering unit is used to perform homogeneous clustering of all risk control messages based on risk control scenarios and intent tags; the feature weight adjustment unit is used to compare homogeneous risk control messages and adjust the feature weights of each risk control feature under different risk control scenarios.

[0096] The risk control early warning and identification module includes a feature weighted analysis unit and an abnormal risk control early warning unit.

[0097] The feature weighted analysis unit is used to extract features and identify scenarios from real-time collected risk control messages, and to generate real-time risk values ​​by weighting each feature based on the risk control scenario; the abnormal risk control early warning unit is used to push early warnings for abnormal risk control messages based on preset risk assessment thresholds.

[0098] 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 invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A data-driven intelligent matching method for risk control messages, characterized in that: The matching method includes the following steps: Step S100: Collect risk control messages and behavioral data from multiple sources to construct several risk control datasets; analyze the behavioral data in the risk control datasets and identify the risk control scenario in which any risk control dataset is located; Step S200: Perform semantic parsing on the risk control messages in any risk control dataset to extract several risk control features; perform combined analysis on the several risk control features to output the intent label corresponding to the risk control dataset; Step S300: Based on risk control scenarios and intent tags, perform homogeneous clustering on all risk control messages; compare homogeneous risk control messages and adjust the feature weights of each risk control feature under different risk control scenarios; Step S400: Extract features and identify scenarios from the real-time collected risk control messages; perform weighted calculations on each feature based on the risk control scenario to generate a real-time risk value; and push early warnings to abnormal risk control messages based on preset risk assessment thresholds. Step S300 includes the following steps: Step S301: Obtain the risk control scenario and intent tag of each risk control message. Randomly select two risk control messages. If the risk control scenario and intent tag of the two risk control messages are the same, divide the two risk control messages into the same type of risk control messages. Divide all risk control messages into several sets of the same type of risk control messages. Step S302: Arbitrarily select a risk control scenario and an intent label, and extract the set of risk control messages of the same type corresponding to the selected risk control scenario and intent label; set all the risk control messages in the extracted set of risk control messages of the same type as target risk control messages, arbitrarily select a target risk control message, extract several risk control features from the selected target risk control message, arbitrarily select one risk control feature from the several risk control features, and count the number of times the selected risk control feature appears in each risk control message as H, where the number of times the selected risk control feature appears in each target risk control message is h1, and calculate the occurrence frequency f of the selected risk control feature under the selected risk control scenario and intent label as f=h1 / H; Step S303: Summarize several risk control features of each target risk control information in the extracted risk control message set to obtain a risk control feature set for selecting risk control scenarios and intent tags. Obtain the occurrence frequency of each risk control feature in the risk control feature set, and set the occurrence frequency of the i-th risk control feature as f. i According to the formula: ； Where e is the number of risk control features in the risk control feature set; the first weight Q1 for selecting risk control features is calculated; Step S304: Extract intent tags containing risk control features in the selected risk control scenario and set them as target intent tags. Count the number of target intent tags and obtain the frequency of occurrence of risk control features in any target intent tag. Set the frequency of occurrence of risk control features in the j1th type of target intent tag as f(j1), according to the formula: ； Where g is the number of target intent tags; the second weight Q2 for selecting risk control features is calculated; Step S305: Obtain the first weight Q1 and the second weight Q2 of the selected risk control feature in the selected risk control scenario, and calculate the feature weight of the selected risk control feature in the selected risk control scenario as Q=Q1×(1+Q2).

2. The intelligent matching method for risk control messages based on data analysis according to claim 1, characterized in that: Step S100 includes the following steps: Step S101: Pre-set several information collection channels to collect risk control messages and user behavior data. At each unit period, summarize the risk control messages and user behavior data collected from all information collection channels, and clean, format and normalize the summarized risk control messages and user behavior data to generate a corresponding risk control dataset. Step S102: Randomly select a risk control dataset, extract user behavior data from the selected risk control dataset, extract corresponding behavioral features from the arbitrary user behavior data, and obtain several behavioral features of the selected risk control dataset; pre-establish a scenario database, which stores several risk control scenarios, in which each risk control scenario is matched with a corresponding scenario feature set; arbitrarily select a risk control scenario, compare the scenario feature set of the selected risk control scenario with several behavioral features of the selected risk control dataset, and if the scenario feature set contains several behavioral features, then set the selected risk control scenario as the risk control scenario in which the selected risk control dataset is located.

3. The intelligent matching method for risk control messages based on data analysis according to claim 2, characterized in that: Step S200 includes the following steps: Step S201: Randomly select a risk control dataset, obtain the risk control messages in the selected risk control dataset, and randomly extract one risk control message from it. Use natural language processing technology to perform text content recognition and word segmentation on the extracted risk control message to obtain several keywords. Step S202: Pre-build a feature database to store several risk control features. Randomly select a keyword from several keywords and compare the similarity between the selected keyword and each risk control feature in terms of word meaning. Obtain the similarity between the selected keyword and each risk control feature. Preset a similarity threshold. If the similarity of several risk control features exceeds the similarity threshold, set the risk control feature with the highest similarity as the risk control feature corresponding to the selected keyword. Obtain the risk control features corresponding to each keyword and generate a risk control feature set. Step S203: Pre-build an intent tag database and store several intent tags. Any intent tag matches a risk control feature group. Randomly select an intent tag and compare the risk control feature set with the risk control feature group of the selected intent tag. If any risk control feature in the risk control feature set belongs to the risk control feature group, then set the selected intent tag as the target intent tag. Step S204: If there are several target intent tags in the risk control feature set, then arbitrarily select one target intent tag, count the number of risk control features in the selected target intent tag as M, obtain the number of risk control features in the risk control feature set as a, and calculate the correlation degree of the selected target intent tag G=a / M; select the target intent tag with the highest correlation degree and set it as the intent tag for extracting risk control messages.

4. The intelligent matching method for risk control messages based on data analysis according to claim 1, characterized in that: Step S400 includes the following steps: Step S401: Whenever a risk control message is collected in real time, semantic parsing and word segmentation are performed on the risk control message to extract several real-time keywords; the semantic similarity of each real-time keyword is compared with the risk control features stored in the feature database, and real-time risk control features with similarity exceeding a preset similarity threshold are selected to form a set of real-time risk control features for the risk control message; the set of real-time risk control features is compared with each intent tag in the intent tag database to obtain the real-time intent tag corresponding to the risk control message. Step S402: Collect user behavior data when risk control messages are collected in real time, extract several behavioral features from the user behavior data, match the several behavioral features with the set of scene features of various risk control scenarios stored in the scene database, and identify the real-time risk control scenario in which the risk control message is located. Step S403: Randomly select the k-th real-time risk control feature from the real-time risk control feature set, and obtain the feature weight Q of the k-th real-time risk control feature under the real-time risk control scenario and real-time intent label. k According to the formula: ； Where s is the number of real-time risk control features; the real-time risk value R of the real-time collected risk control messages is calculated; and a risk assessment threshold R is preset. th If R≥R th If so, the risk control message will be marked as abnormal and a risk warning will be pushed.

5. A risk control message intelligent matching system, used to execute the risk control message intelligent matching method based on data analysis according to any one of claims 1-4, characterized in that: The matching system includes a risk control data analysis module, a risk intent analysis module, a risk weighting module, and a risk control early warning identification module. The risk control data analysis module is used to collect risk control messages and behavioral data from multiple sources to construct several risk control datasets; analyze the behavioral data in the risk control datasets; and identify the risk control scenario in which any risk control dataset is located. The risk intent analysis module is used to perform semantic parsing on risk control messages in any risk control dataset and extract several risk control features. Combine and analyze several risk control features to output the intent label corresponding to the risk control dataset; The risk weighting module is used to perform homogeneous clustering on all risk control messages based on risk control scenarios and intent tags; compare homogeneous risk control messages, and adjust the feature weights of each risk control feature under different risk control scenarios; The risk control early warning and identification module is used to extract features and identify scenarios from real-time collected risk control messages, and to generate real-time risk values ​​by weighting each feature based on the risk control scenario. Based on preset risk assessment thresholds, early warnings are pushed out for abnormal risk control messages.

6. The intelligent matching system for risk control messages according to claim 5, characterized in that: The risk control data analysis module includes a risk control data acquisition unit and a risk control scenario recognition unit; The risk control data acquisition unit is used to collect risk control messages and behavioral data from multiple sources to construct several risk control datasets; the risk control scenario identification unit is used to analyze the behavioral data in the risk control datasets and identify the risk control scenario in which any risk control dataset is located.

7. The intelligent matching system for risk control messages according to claim 5, characterized in that: The risk intent analysis module includes a risk control message parsing unit and a risk intent identification unit; The risk control message parsing unit is used to perform semantic parsing on risk control messages in any risk control dataset and extract several risk control features; the risk intent recognition unit is used to perform combined analysis on several risk control features and output intent labels corresponding to the risk control dataset.

8. The intelligent matching system for risk control messages according to claim 5, characterized in that: The risk weighting module includes a risk control message clustering unit and a feature weighting adjustment unit; The risk control message clustering unit is used to perform homogeneous clustering of all risk control messages based on risk control scenarios and intent tags; the feature weight adjustment unit is used to compare homogeneous risk control messages and adjust the feature weights of each risk control feature under different risk control scenarios.

9. The intelligent matching system for risk control messages according to claim 5, characterized in that: The risk control early warning identification module includes a feature weighted analysis unit and an abnormal risk control early warning unit; The feature weighting analysis unit is used to extract features and identify scenarios from the real-time collected risk control messages, and to generate real-time risk values ​​by weighting each feature based on the risk control scenario; the abnormal risk control early warning unit is used to push early warnings to abnormal risk control messages based on preset risk assessment thresholds.

Images