ChatGLM model-based risk point automatic detection and identification system and method
By combining the ChatGLM model and local risk control audit experience knowledge base, an automatic risk point detection and identification system was built, which solved the problem of untimely and inaccurate risk identification in the existing technology, and achieved efficient and accurate risk identification and intelligent decision-making support .
Patent Information
- Application Number
- CN202411156122.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-22
- Publication Date
- 2025-01-14
AI Technical Summary
In the existing technology, risk identification relies on manual review and experience knowledge base, which has problems such as slow processing speed, low recognition accuracy, lagging knowledge base updates, lack of intelligent assisted decision-making and semantic understanding capabilities, making it difficult to adapt to large-scale data analysis and new types. Risk identification.
Adopt the automatic risk point detection and identification system based on the ChatGLM model, and through the combination of semantic understanding rule base and local risk control audit experience knowledge base, the automatic risk point identification module and auxiliary decision-making opinion module are built, and risk characteristics are extracted and auxiliary decision-making is generated using the ChatGLM model. Opinions to achieve intelligent decision-making support.
It improves the efficiency and accuracy of risk identification, reduces the cumbersomeness and misjudgment of manual review, can respond to new risks and complex situations in a timely manner, and enhances semantic understanding ability.
Smart Images
Figure CN119312154A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence, and in particular to a system and method for automatically detecting and identifying risk points based on a ChatGLM model. Background Art
[0002] With the rapid development of information technology, technologies such as big data and artificial intelligence have been widely used in finance, e-commerce and other fields. However, the automatic detection and identification of risk points has always been a difficult problem faced by these fields. Although the traditional local risk control and audit experience knowledge base has accumulated a lot of experience, it lacks real-time updating and intelligent processing capabilities.
[0003] In the existing technology, risk identification mainly relies on manual review and experience knowledge base, which has the following obvious defects: First, the processing speed is slow and cannot adapt to the efficient analysis of large-scale data; second, the recognition accuracy is low, and it is easy to miss or misjudge risk points; third, the knowledge base is updated lagging behind, making it difficult to deal with new risks; fourth, there is a lack of intelligent decision-making support, and the review process is cumbersome and inefficient; fifth, the semantic understanding ability is limited, and the ability to recognize complex situations and risk patterns is not strong.
[0004] The combination of the semantic understanding capability of the big model and the local risk control and audit experience knowledge base can optimize the above problems, improve the efficiency and accuracy of risk identification, and realize intelligent decision-making support, which is in line with the development trend of modern technology. Summary of the invention
[0005] In view of this, the present application provides a risk point automatic detection and identification system and method based on the ChatGLM model to achieve automatic detection and identification of risk points and provide auxiliary decision-making opinions, thereby improving the efficiency and accuracy of risk control.
[0006] The first aspect of the present application provides a risk point automatic detection and identification system based on the ChatGLM model, the system comprising a semantic understanding rule base construction module, a local risk control audit experience knowledge base construction module, a risk point automatic identification module and an auxiliary decision-making opinion module;
[0007] The semantic understanding rule base construction module is used to process the collected corpus data to obtain training data, and train the pre-built ChatGLM model through the training data to obtain the semantic understanding rule base;
[0008] The local risk control review experience knowledge base construction module is used to screen, classify and organize the collected historical risk control related information to generate target risk control information; the local risk control review experience knowledge base is generated through the target risk control information according to the preset local risk control review experience knowledge base structure;
[0009] The risk point automatic identification module is used to determine the risk point data and features in the local risk control and audit experience knowledge base and mark them; and compare them with the target risk points and features extracted by the ChatGLM model, determine the risk point results of the target risk points and send them to the auxiliary decision-making opinion module;
[0010] The auxiliary decision opinion module is used to generate auxiliary decision opinions for the risk points after receiving the risk points identified by the risk point automatic identification module by retrieving the target risk control information in the local risk control audit experience knowledge base and the rules in the semantic understanding rule base.
[0011] Optionally, the processing of the collected corpus data to obtain training data includes:
[0012] Performing data cleaning on the corpus data to remove special characters, punctuation marks, stop words and invalid data in the corpus data to generate preprocessed corpus data;
[0013] Performing word segmentation and feature extraction on the preprocessed corpus data, and dividing the extracted features into a training set, a validation set, and a test set;
[0014] The training data is generated by using the training set, validation set and test set.
[0015] Optionally, extracting the risk points of the local risk control and audit experience knowledge base and the description and solution of the risk points includes:
[0016] Extract target features related to risk points based on historical data and risk control review cases;
[0017] Input the target feature into the trained ChatGLM model, determine the risk point keywords of the target feature, and determine the risk points of the local risk control audit experience knowledge base according to the risk point keywords;
[0018] The type of risk point is determined through a classification algorithm, and the description and solution of the risk point are determined by combining the rules and cases in the local risk control and audit experience knowledge base.
[0019] Optionally, the system further includes an automatic update module:
[0020] The automatic update module is used to regularly collect target corpus data for updating, and to review and update the semantic understanding rule base through the target corpus data;
[0021] Regularly collect target risk control data for updating, and review and update the local risk control audit experience knowledge base through the target risk control data;
[0022] Collect user feedback and actual implementation status on the auxiliary decision-making opinions, and optimize the auxiliary decision-making opinion module based on the feedback and actual implementation status.
[0023] Optionally, the historical risk control related information includes past risk control review cases, policy documents, industry standards and relevant regulations.
[0024] The second aspect of the present application provides a method for automatically detecting and identifying risk points based on a ChatGLM model, the method comprising:
[0025] Processing the collected corpus data to obtain training data, training the pre-built ChatGLM model with the training data to generate a semantic understanding rule base;
[0026] Screen, classify and organize the collected historical risk control related information to generate target risk control information, and generate the local risk control review experience knowledge base through the target risk control information according to the preset local risk control review experience knowledge base structure;
[0027] Receive the audit data file, determine the risk points of the audit data file and auxiliary decision-making opinions for the risk points, and send the auxiliary decision-making opinions to the risk control auditor.
[0028] Optionally, the processing of the collected corpus data to obtain training data includes:
[0029] Performing data cleaning on the corpus data to remove special characters, punctuation marks, stop words and invalid data in the corpus data to generate preprocessed corpus data;
[0030] Performing word segmentation and feature extraction on the preprocessed corpus data, and dividing the extracted features into a training set, a validation set, and a test set;
[0031] The training data is generated by using the training set, validation set and test set.
[0032] Optionally, the determining of the risk points of the audit data file and the auxiliary decision-making opinions of the risk points includes:
[0033] Using the ChatGLM model to read the audit data file related to risk control, parsing the audit data file into a number of content data paragraphs, retrieving keywords and key paragraphs of the content data paragraphs through the semantic understanding rule library, and determining the risk point data of the audit data file according to the keywords and key paragraphs;
[0034] According to the features and risk point data marked in the local risk control and audit experience knowledge base, the risk points of the audit data file are determined by comparing them with the risk point data extracted by the ChatGLM model, and auxiliary decision-making opinions on the risk points are determined through the local risk control and audit experience knowledge base.
[0035] Optionally, after generating the semantic understanding rule base, the method further includes:
[0036] regularly collecting target corpus data for updating, and reviewing and updating the semantic understanding rule base through the target corpus data;
[0037] After generating the local risk control review experience knowledge base, the method further includes:
[0038] Regularly collect target risk control data for updating, and review and update the local risk control audit experience knowledge base through the target risk control data.
[0039] Optionally, the historical risk control related information includes past risk control review cases, policy documents, industry standards and relevant regulations.
[0040] In the embodiment provided in the present application, a semantic understanding rule base is constructed through corpus data, and then a local risk control audit experience knowledge base is constructed after processing historical risk control related materials, and the rule base, knowledge base and ChatGLM model are combined, so that when risk points are identified through the ChatGLM model, the model can use the risk control rules and experience knowledge in the rule base and knowledge base to improve the accuracy and efficiency of identification. And based on the semantic understanding results of the ChatGLM model, combined with the knowledge base, auxiliary decision-making opinions on risk points can be automatically generated, providing intelligent support for auditors, effectively reducing the cumbersome, untimely and misjudgment problems of manual audits. The present invention solves the problem of untimely and inaccurate risk identification in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 A system module diagram provided for an embodiment of the present application;
[0042] Figure 2 A flow chart of the method provided in the embodiment of the present application;
[0043] Figure 3 A schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0044] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.
[0045] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.
[0046] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".
[0047] The present application provides a risk point automatic detection and identification system and method based on the ChatGLM model to solve the problem of untimely and inaccurate risk identification in the prior art.
[0048] like Figure 1 As shown, this is a system module diagram of a risk point automatic detection and identification system based on the ChatGLM model provided by the present application. The system includes: a semantic understanding rule base construction module, a local risk control audit experience knowledge base construction module, a risk point automatic identification module and an auxiliary decision-making opinion module.
[0049] The functions of each module are described below:
[0050] 1. Semantic understanding rule base construction module. This module is used to build a semantic understanding rule base.
[0051] First, we can collect corpus data of semantic rule base from various online corpora, such as Fahai Risk Control, SmoothNLP, and People's Data. Corpus data can include text, conversation records, etc. By extracting features from the corpus data, we can obtain training data for model training. After inputting the training data into ChatGLM model training, we can complete the construction of structured semantic understanding rule base.
[0052] And after the rule base is constructed, an API interface for iterative updating of the rule base can be provided to the outside. Through this API interface, the latest corpus data can be received regularly, and the rule base can be updated through this latest corpus data.
[0053] In another embodiment, the processing of the collected corpus data to obtain training data includes:
[0054] Performing data cleaning on the corpus data to remove special characters, punctuation marks, stop words, and invalid data in the corpus data, and generating preprocessed corpus data;
[0055] Performing word segmentation and feature extraction on the preprocessed corpus data, and dividing the extracted features into a training set, a validation set, and a test set;
[0056] Generating the training data through the training set, the validation set, and the test set.
[0057] In this embodiment, since there will be a large amount of noise and invalid data in the collected corpus data, such as special characters, HTML tags, punctuation, missing values, and some stop words, such as "de", "he", "shi", etc. Therefore, through preprocessing, the above-mentioned invalid data can be cleaned, thereby reducing some invalid operations and improving the efficiency of model training. And dividing the preprocessed corpus data into a training set, a validation set, and a test set can facilitate model training more conveniently.
[0058] 2. Local risk control review experience knowledge base construction module. This module is used to screen, classify, and organize the collected historical risk control-related materials to generate target risk control materials; generate the local risk control review experience knowledge base through the target risk control materials according to the preset local risk control review experience knowledge base structure.
[0059] In this module, first, the overall structure of the knowledge base, including the directory, classification, and index, needs to be determined before constructing the local risk control review experience knowledge base. So that when classifying and organizing the historical risk control-related materials subsequently, it can be carried out according to the structure of this knowledge base. The historical risk control-related materials can include risk control review cases, policy documents, industry standards, and relevant regulations, etc.
[0060] When starting to build, you can collect the above information through official government websites, industry associations and standardization organizations, professional legal databases, corporate official websites, risk control service providers and other platforms. And in order to ensure the accuracy and completeness of the information, first screen it and remove invalid data in the information. Then classify and organize it according to the structure of the knowledge base, so as to obtain the target risk control information with the same structure as the knowledge base. After the target risk control information is added to the knowledge base according to the structure, the construction of the local risk control audit experience knowledge base can be completed.
[0061] 3. Automatic risk point identification module. This module is used to determine the risk point data and features in the local risk control and audit experience knowledge base and mark them; and compare them with the target risk points and features extracted by the ChatGLM model, determine the risk point results of the target risk points and send them to the auxiliary decision-making opinion module.
[0062] In this module, the local risk control audit experience knowledge base is first preprocessed to extract the various risk points in the knowledge base, as well as the descriptions and solutions of each risk point. Then, based on the historical data and risk control audit cases in the knowledge base, the features related to the risk points are extracted and annotated, such as transaction amount, frequency, user behavior, etc. When the audit data sent by the ChatGLM model is received, the new audit data is generated by comparing the above-annotated features and risk point data, and then input into the ChatGLM model to automatically identify potential risk point keywords, and attribute them to the corresponding risk point types through the classification algorithm. Combined with the rules and cases in the knowledge base, detailed analysis and judgment of the risk points are performed to generate risk point results, which can include several risk points and corresponding analysis results. And the risk point results can be fed back to the risk control auditors for their reference and verification.
[0063] In this module, similarity matching is used for keyword risk points, scenarios, and descriptions extracted from the local risk control and audit experience knowledge base, which greatly improves the efficiency of risk identification and improves the accuracy of identification.
[0064] 4. Decision-making assistance opinion module. This module is used to provide decision-making assistance opinions for each risk point after receiving the risk point results identified by the risk point automatic identification module, such as risk disposal suggestions, approval strategy adjustments, risk control measures, etc., by retrieving the target risk control information in the local risk control review experience knowledge base and the rules in the semantic understanding rule base, so as to help decision makers make more informed decisions.
[0065] So far, completed Figure 1 Functional description of each module in .
[0066] In the embodiment of the present application, a semantic understanding rule base is constructed through corpus data, and then a local risk control audit experience knowledge base is constructed after processing historical risk control related materials, and the rule base, knowledge base and ChatGLM model are combined, so that when risk points are identified through the ChatGLM model, the model can use the risk control rules and experience knowledge in the rule base and knowledge base to improve the accuracy and efficiency of identification. And based on the semantic understanding results of the ChatGLM model, combined with the knowledge base, auxiliary decision-making opinions on risk points can be automatically generated, providing intelligent support for auditors, effectively reducing the tediousness of manual audits, untimely and misjudgment problems. The present invention solves the problem of untimely and inaccurate risk identification in the prior art.
[0067] In another embodiment, the above system further includes an automatic update module:
[0068] The automatic update module is used to regularly collect target corpus data for updating, and to review and update the semantic understanding rule base through the target corpus data;
[0069] Regularly collect target risk control data for updating, and review and update the local risk control audit experience knowledge base through the target risk control data;
[0070] Collect user feedback and actual implementation status on the auxiliary decision-making opinions, and optimize the auxiliary decision-making opinion module based on the feedback and actual implementation status.
[0071] In this module, the local risk control and audit experience knowledge base can identify new risks and reduce missed or misjudged risk points through regular review and updating. After collecting feedback and actual implementation of auxiliary decision-making opinions, the processing solutions in the auxiliary decision-making opinions can be scored, and the processing solutions with scores higher than the threshold can be added to the local risk control and audit experience knowledge base. This will further optimize the local risk control and audit experience knowledge base and improve semantic understanding capabilities.
[0072] like Figure 2 As shown, the present application also provides a method for automatically detecting and identifying risk points based on the ChatGLM model, the method comprising:
[0073] Step S201, processing the collected corpus data to obtain training data, training the pre-built ChatGLM model with the training data, and generating a semantic understanding rule base.
[0074] Step S202, screen, classify and organize the collected historical risk control related information to generate target risk control information, and generate the local risk control audit experience knowledge base through the target risk control information according to the preset local risk control audit experience knowledge base structure.
[0075] Step S203, receiving the audit data file, determining the risk points of the audit data file and auxiliary decision-making opinions of the risk points, and sending the auxiliary decision-making opinions to the risk control auditor.
[0076] In another embodiment, the processing of the collected corpus data to obtain training data includes:
[0077] Performing data cleaning on the corpus data to remove special characters, punctuation marks, stop words and invalid data in the corpus data to generate preprocessed corpus data;
[0078] Performing word segmentation and feature extraction on the preprocessed corpus data, and dividing the extracted features into a training set, a validation set, and a test set;
[0079] The training data is generated by using the training set, validation set and test set.
[0080] In another embodiment, the determining of the risk points of the audit data file and the auxiliary decision-making opinions of the risk points includes:
[0081] Using the ChatGLM model to read the audit data file related to risk control, parsing the audit data file into a number of content data paragraphs, retrieving keywords and key paragraphs of the content data paragraphs through the semantic understanding rule library, and determining the risk point data of the audit data file according to the keywords and key paragraphs;
[0082] According to the features and risk point data marked in the local risk control and audit experience knowledge base, the risk points of the audit data file are determined by comparing them with the risk point data extracted by the ChatGLM model, and auxiliary decision-making opinions on the risk points are determined through the local risk control and audit experience knowledge base.
[0083] In another embodiment, after generating the semantic understanding rule base, the method further includes:
[0084] regularly collecting target corpus data for updating, and reviewing and updating the semantic understanding rule base through the target corpus data;
[0085] After generating the local risk control review experience knowledge base, the method further includes:
[0086] Regularly collect target risk control data for updating, and review and update the local risk control audit experience knowledge base through the target risk control data.
[0087] In another embodiment, the historical risk control related information includes past risk control review cases, policy documents, industry standards and relevant regulations.
[0088] The above embodiment of the present invention provides a risk point automatic detection and identification system based on the ChatGLM model, and based on the system provides a risk point automatic detection and identification method based on the ChatGLM model. Through the above system and method, the problem of untimely and inaccurate risk identification in the prior art is solved.
[0089] This embodiment also discloses a computer device, such as Figure 3 As shown, the computer device includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement any of the above-mentioned automatic detection and identification methods for risk points based on the ChatGLM model.
[0090] In addition, in the above-mentioned example implementation of the automatic detection and identification system for risk points based on the ChatGLM model, the logical division of each program module is only an example. In actual applications, the above-mentioned functions can be assigned to different program modules as needed, for example, for the convenience of corresponding hardware configuration requirements or software implementation. That is, the internal structure of the automatic detection and identification system for risk points based on the ChatGLM model is divided into different program modules to complete all or part of the functions described above.
[0091] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.
Claims
1. A risk point automatic detection and identification system based on ChatGLM model, characterized in that: The system includes a semantic understanding rule base construction module, a local risk control review experience knowledge base construction module, a risk point automatic identification module and an auxiliary decision-making opinion module; The semantic understanding rule base construction module is used to process the collected corpus data to obtain training data, and train the pre-built ChatGLM model through the training data to obtain the semantic understanding rule base; The local risk control review experience knowledge base construction module is used to screen, classify and organize the collected historical risk control related information to generate target risk control information; the local risk control review experience knowledge base is generated through the target risk control information according to the preset local risk control review experience knowledge base structure; The risk point automatic identification module is used to determine the risk point data and features in the local risk control and audit experience knowledge base and mark them; and compare the target risk points and features extracted by the ChatGLM model to determine the risk point results of the target risk points and send them to the auxiliary decision-making opinion module; The auxiliary decision opinion module is used to generate auxiliary decision opinions for the risk points after receiving the risk points identified by the risk point automatic identification module by retrieving the target risk control information in the local risk control audit experience knowledge base and the rules in the semantic understanding rule base.
2. The system according to claim 1, characterized in that The step of processing the collected corpus data to obtain training data includes: Performing data cleaning on the corpus data to remove special characters, punctuation marks, stop words and invalid data in the corpus data to generate preprocessed corpus data; Performing word segmentation and feature extraction on the preprocessed corpus data, and dividing the extracted features into a training set, a validation set, and a test set; The training data is generated by using the training set, validation set and test set.
3. The system according to claim 1, characterized in that The extracting of risk points from the local risk control and audit experience knowledge base and the description and solution of the risk points includes: Extract target features related to risk points based on historical data and risk control review cases; Input the target feature into the trained ChatGLM model, determine the risk point keywords of the target feature, and determine the risk points of the local risk control audit experience knowledge base according to the risk point keywords; The type of risk point is determined through a classification algorithm, and the description and solution of the risk point are determined by combining the rules and cases in the local risk control and audit experience knowledge base.
4. The system according to claim 1, characterized in that The system also includes an automatic update module: The automatic update module is used to regularly collect target corpus data for updating, and to review and update the semantic understanding rule base through the target corpus data; Regularly collect target risk control data for updating, and review and update the local risk control audit experience knowledge base through the target risk control data; Collect user feedback and actual implementation status on the auxiliary decision-making opinions, and optimize the auxiliary decision-making opinion module based on the feedback and actual implementation status.
5. The system according to claim 1, characterized in that The historical risk control related information includes past risk control review cases, policy documents, industry standards and relevant regulations.
6. A method for automatically detecting and identifying risk points based on the ChatGLM model, characterized in that: The method comprises: Processing the collected corpus data to obtain training data, training the pre-built ChatGLM model with the training data to generate a semantic understanding rule base; Screen, classify and organize the collected historical risk control related information to generate target risk control information, and generate the local risk control review experience knowledge base through the target risk control information according to the preset local risk control review experience knowledge base structure; Receive the audit data file, determine the risk points of the audit data file and auxiliary decision-making opinions for the risk points, and send the auxiliary decision-making opinions to the risk control auditor.
7. The method according to claim 6, characterized in that The step of processing the collected corpus data to obtain training data includes: Performing data cleaning on the corpus data to remove special characters, punctuation marks, stop words and invalid data in the corpus data to generate preprocessed corpus data; Performing word segmentation and feature extraction on the preprocessed corpus data, and dividing the extracted features into a training set, a validation set, and a test set; The training data is generated by using the training set, validation set and test set.
8. The method according to claim 6, characterized in that The determination of the risk points of the audit data file and the auxiliary decision-making opinions of the risk points include: Using the ChatGLM model to read the audit data file related to risk control, parsing the audit data file into a number of content data paragraphs, retrieving keywords and key paragraphs of the content data paragraphs through the semantic understanding rule library, and determining the risk point data of the audit data file according to the keywords and key paragraphs; According to the features and risk point data marked in the local risk control and audit experience knowledge base, the risk points of the audit data file are determined by comparing them with the risk point data extracted by the ChatGLM model, and auxiliary decision-making opinions on the risk points are determined through the local risk control and audit experience knowledge base.
9. The method according to claim 6, characterized in that After generating the semantic understanding rule base, the method further includes: regularly collecting target corpus data for updating, and reviewing and updating the semantic understanding rule base through the target corpus data; After generating the local risk control review experience knowledge base, the method further includes: Regularly collect target risk control data for updating, and review and update the local risk control audit experience knowledge base through the target risk control data.
10. The method according to claim 6, characterized in that The historical risk control related information includes past risk control review cases, policy documents, industry standards and relevant regulations.
Citation Information
Cited By
Radio management system and method
CN120430758A