A credit risk monitoring method for dishonest subject behavior

Abstract

The application discloses a credit risk monitoring method for discredited subject behavior, comprising the following steps: obtaining discredited event data; obtaining basic types of discredited events according to the clustering of the discredited event data; determining the category label and keyword corpus of discredited public opinion according to the basic types of discredited events; training an LDA topic model by using the keywords in the keyword corpus; processing the keywords in the keyword corpus and storing them in a database; cleaning the discredited public opinion information in the database; performing natural language processing on the cleaned discredited public opinion information to generate intelligent labels. The application applies the constructed credit keyword library and machine learning crawler to intelligently crawl relevant discredited public opinion information, thereby improving the precision and breadth of the discredited public opinion information crawling. The data quality platform is used to clean the public opinion information, and the NLP text processing technology is used to generate intelligent labels required by various public opinions from the cleaned information, so that the discredited public opinion information can be quickly and accurately analyzed.

Description

Technical Field

[0001] This invention relates to the field of credit risk monitoring technology, and in particular to a credit risk monitoring method targeting the behavior of defaulting entities. Background Technology

[0002] Internet credit information acquisition and analysis is a research field that has developed in recent years with the advancement of databases and networks. It mainly includes the acquisition, storage, cleaning, and analysis of credit information. As a crucial basis for government regulatory departments to assess the risk of credit default, public opinion regarding credit defaults needs to be acquired as broadly as possible regarding credit entities in specific fields, industries, and regions, in order to take appropriate action. The processing of public opinion information on credit defaults includes: the category of credit default in the relevant public opinion information, its geographical location, sentiment analysis, keywords, abstracts, dissemination paths, and media outlets.

[0003] Currently, the collection of information on public opinion regarding breaches of trust largely relies on traditional methods. This involves crawling information using keywords provided by domain experts, storing it in a database, cleaning the information, and then performing sentiment analysis. However, for information on breaches of trust, the limited number of domain keywords provided by experts often leads to inaccurate information. Furthermore, simple sentiment analysis is insufficient to meet the requirements of credit regulatory departments regarding public opinion on breaches of trust.

[0004] Therefore, how to effectively, quickly, and accurately capture specific news events related to breaches of trust, and how to quickly generate information such as sentiment, domain, region, news summary, and keywords, are urgent problems to be solved in this field. Summary of the Invention

[0005] This invention provides a credit risk monitoring method for entities that have breached trust, in order to solve the problems in the prior art where the capture of public opinion information on breaches of trust is inaccurate and the analysis of such information is insufficient to meet regulatory requirements.

[0006] On the one hand, embodiments of the present invention provide a credit risk monitoring method for the behavior of entities with poor credit records, including:

[0007] Obtain data on breaches of trust;

[0008] The basic types of breaches of trust are obtained by clustering the data on breaches of trust.

[0009] Based on the basic types of breaches of trust, determine the category tags and corresponding keyword corpus of public opinion regarding breaches of trust;

[0010] LDA topic model training was conducted using keywords from a keyword corpus to expand the keyword corpus;

[0011] The keywords in the keyword corpus are processed and stored in the database;

[0012] The information on public opinion regarding dishonesty in the database is imported into the data quality processing platform, which then cleans the information.

[0013] Natural language processing is used to generate intelligent tags from the cleaned-up information on public opinion regarding breaches of trust.

[0014] The credit risk monitoring method for entities that have breached their creditworthiness, as described in this invention, has the following advantages:

[0015] The application utilizes a built-in credit keyword database and machine learning crawlers to intelligently crawl relevant information on public opinion regarding dishonesty, improving the accuracy and breadth of this information. A data quality platform is used to clean the public opinion information, and NLP text processing technology is employed to generate various intelligent tags for different public opinion topics, achieving rapid and accurate analysis of information on dishonesty. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a credit risk monitoring method for entities that have breached their creditworthiness, provided as an embodiment of the present invention. Detailed Implementation

[0018] 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.

[0019] Figure 1 This invention provides a flowchart of a credit risk monitoring method targeting the behavior of entities with poor credit records. The invention provides a credit risk monitoring method targeting the behavior of entities with poor credit records, including:

[0020] S100, Obtain data on breaches of trust.

[0021] For example, data on breaches of trust can be obtained from the credit China breaches of trust event database, which records various breaches of trust events, including information such as the subject of the breach of trust punishment, the basis for the punishment, the punishment result, and the punishment event.

[0022] S110. Obtain the basic types of dishonesty events by clustering the data on dishonesty events.

[0023] For example, the basic types of breach of trust events can be obtained by clustering using the jieba and K-means algorithms.

[0024] S120. Determine the category labels and corresponding keyword corpus of public opinion on dishonesty based on the basic types of dishonesty incidents.

[0025] For example, the determined category labels should meet the following requirements: cover as many different grounds for punishment as possible, have good distinguishability between category labels, and each category label corresponds to a keyword corpus for that category. Initially, category labels can be extracted manually based on the specific textual information of the punishment, such as categories for unpaid wages and unpaid taxes.

[0026] S130. Use keywords from the keyword corpus to train an LDA topic model to expand the keyword corpus.

[0027] For example, LDA is short for Latent Dirichlet Allocation, which is a topic model used to represent the topics of a document in the form of a probability distribution.

[0028] S130 specifically includes: using keywords in the keyword corpus and crawling relevant web pages using web crawling tools; extracting the main text or abstract from the web pages; using the extracted main text or abstract as a new keyword corpus for LDA, and updating the topic by combining it with the original keyword corpus.

[0029] After obtaining the main text or summary of a webpage, a semantic similarity calculation model can be used to determine the semantic similarity of short texts within the webpage. This guides web crawling tools to improve the coverage of crawled data while maintaining accuracy and efficiency. Webpages crawled by web crawling tools often exhibit topic bias, so the keyword corpus expanded based on these crawled webpages will also show a bias towards a certain topic. However, after training with the LDA topic model in this invention, topic documents can avoid the original topic, resulting in better granularity and coverage.

[0030] S140. Process the keywords in the keyword corpus and store them in the database.

[0031] For example, keyword processing includes converting the keyword format into a specific format, which includes setting fields, ID, title, content, time, source, etc. The processed data can be stored in a MySQL database.

[0032] S150. Import the information on public opinion regarding dishonesty from the database into the data quality processing platform, and have the data quality processing platform clean the information on public opinion regarding dishonesty.

[0033] For example, the data quality processing platform cleans up information on public opinion that has lost credibility by performing the following processes: URL deduplication, date formatting, information splitting, information merging, null value filtering, type conversion, and field mapping.

[0034] S160. Perform natural language processing on the cleaned information on public opinion regarding dishonesty to generate intelligent tags.

[0035] For example, smart tags include geographic tags, domain tags, sentiment tags, keyword cloud tags, and summary tags.

[0036] Specifically, the cleaned information on public opinion regarding misconduct can be input into a tag extraction model to obtain regional tags. The cleaned information can be trained using a FastText model and a corpus of pre-labeled keywords to obtain domain tags. Simultaneously, a sentiment analysis model can be used to perform sentiment analysis, keyword cloud extraction, and summary generation on documents within the cleaned information on misconduct, obtaining sentiment tags, keyword cloud tags, and summary tags.

[0037] The aforementioned label extraction model can be built based on RNN (Recurrent Neural Network) and CRF (Conditional Random Field) models, while the sentiment analysis model can use the snowNLP model. snowNLP is a Python library for Chinese Natural Language Processing (NLP) that supports Chinese NLP operations including Chinese word segmentation, part-of-speech tagging, sentiment classification, and text classification.

[0038] S170. Cluster the cleaned information on public opinion regarding breaches of trust, and conduct time evolution analysis and media popularity analysis based on the clustering results.

[0039] For example, time evolution analysis and media popularity analysis are performed based on the clustering results, including: statistical analysis of public opinion information on dishonesty in the same category according to information such as reporting time, reporting media, and number of reports, and generating corresponding statistical charts.

[0040] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0041] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A credit risk monitoring method targeting the behavior of entities with poor credit records, characterized in that, include: The data on breaches of trust are obtained from the database of breaches of trust in Credit China. The data on breaches of trust includes the subject of the breach of trust, the basis for the penalty, the penalty result, and the penalty event. The basic types of dishonesty events are obtained by clustering the data on dishonesty events. Based on the basic types of the breach of trust incidents, determine the category tags and corresponding keyword corpus of public opinion regarding breaches of trust; The category tags are extracted based on the specific text information of the penalty, including categories such as unpaid wages and unpaid taxes. The category tags are distinguishable from each other, and each category tag corresponds to a category of keyword corpus. LDA topic model training is performed using keywords from the keyword corpus to expand the keyword corpus. This includes: using keywords from the keyword corpus and web crawling tools to crawl relevant web pages; extracting the main text or abstract from the web pages; using the extracted main text or abstract as a new keyword corpus for LDA, and updating the topics by combining it with the original keyword corpus; after obtaining the main text or abstract from the web pages, using a semantic similarity calculation model to determine the semantic similarity of short texts in the web pages, guiding the web crawling tools to improve the coverage of crawled data, while taking into account the accuracy and efficiency of data crawling. The keywords in the keyword corpus are processed and stored in the database; The information on public opinion regarding dishonesty in the database is imported into the data quality processing platform, which then cleans the information. Natural language processing is performed on the cleaned public opinion information regarding breaches of trust to generate intelligent tags. These intelligent tags include regional tags, domain tags, sentiment tags, keyword cloud tags, and summary tags. The generation of intelligent tags includes: inputting the cleaned public opinion information regarding breaches of trust into a tag extraction model to obtain regional tags; training the cleaned public opinion information regarding breaches of trust using a FastText model and a keyword corpus with pre-labeled categories to obtain domain tags; and using a sentiment analysis model to perform sentiment analysis, keyword cloud extraction, and summary generation on the documents in the cleaned public opinion information regarding breaches of trust.

2. The credit risk monitoring method for entities with poor credit records according to claim 1, characterized in that, Also includes: The cleaned information on public opinion regarding breaches of trust is then clustered. Based on the clustering results, time evolution analysis and media popularity analysis were performed.

3. The credit risk monitoring method for entities with poor credit records according to claim 2, characterized in that, The time evolution analysis and media popularity analysis based on the clustering results include: For the same category of public opinion information on dishonesty, statistics are compiled according to the reporting time, the reporting media, and the number of reports, and corresponding statistical charts are generated.

4. The credit risk monitoring method for entities with poor credit records according to claim 1, characterized in that, The basic types of dishonesty events obtained by clustering the dishonesty event data include: The data on the breach of trust were clustered using jieba word segmentation and the K-means algorithm to obtain the basic types of the breach of trust events.

5. A credit risk monitoring method for entities with poor credit records according to claim 1, characterized in that, The data quality processing platform cleans the information related to public opinion breaches of trust, including: The aforementioned information on public opinion regarding breaches of trust is processed by URL deduplication, date formatting, information splitting, information merging, null value filtering, type conversion, and field mapping.

6. A credit risk monitoring method for entities with poor credit records according to claim 1, characterized in that, The label extraction model is based on RNN and CRF, and the sentiment analysis model is the snowNLP model.

Images