Enterprise credit assessment method based on semi-supervised learning and related equipment thereof

Abstract

The invention provides an enterprise credit assessment method based on semi-supervised learning and related equipment thereof, and the method comprises the steps: classifying historical loan enterprises into a plurality of categories; for each category, loan default enterprises and high-quality loan enterprises are selected as representative enterprises, data of the enterprises are marked respectively, and a positive sample data set and a negative sample data set are constructed. And training an initial classifier based on the annotated data and the default risk events, and performing iterative training on unannotated data in the categories to gradually improve the enterprise credit assessment classifier of each category. And then, integrating the classifiers of a plurality of categories into one enterprise credit evaluation model. And finally, inputting target loan enterprise data, and generating a credit evaluation result of the enterprise through the evaluation model. According to the enterprise credit assessment scheme based on semi-supervised learning, a small amount of annotated data and a large amount of unannotated data are utilized, so that the assessment capability of the model on the enterprise credit risk is remarkably improved, and reduction of the credit risk and optimization of resource configuration are facilitated.

Description

Technical Field

[0001] The present application relates to the field of science and technology finance technology, and specifically to a corporate credit assessment method based on semi-supervised learning and related equipment. Background Art

[0002] Corporate credit loans refer to a type of loan issued by banks or other financial institutions based on the business conditions, credit record and repayment ability of the enterprise, without the need for physical collateral. They are mainly used to solve the short-term capital needs of enterprises, promote the turnover of corporate funds, and help enterprises develop rapidly. They are characterized by convenient application and fast approval. Corporate credit loans are closely related to corporate credit. A good corporate credit record can often help companies obtain loan approval more easily, enjoy lower interest rates and more flexible repayment terms. On the contrary, companies with poor credit may face difficulties in obtaining loans and high interest rates, which will affect the company's capital operation and development plans.

[0003] Therefore, a key link in corporate credit loans is corporate credit assessment. At present, the main methods of corporate credit assessment include manual assessment based on expert experience and supervised assessment based on machine learning, but both assessment methods have corresponding defects. For example, manual assessment relies on the experience and judgment of the assessor, which may be subjective and inconsistent; while supervised machine learning assessment relies on the accuracy and completeness of historical data. If there is a deviation in data annotation, the prediction result may be inaccurate, and the construction of labeled samples that meet the quantity requirements of the supervised assessment method requires a lot of manual input, which is time-consuming and labor-intensive. In addition, if the data set is not updated in a timely manner, the supervised assessment method may not be able to effectively identify new risks or abnormal data. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a corporate credit assessment method, device, computer equipment and storage medium based on semi-supervised learning to solve the technical problems of existing corporate credit assessment methods, such as large workload and low assessment accuracy.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a corporate credit assessment method based on semi-supervised learning, which adopts the following technical solution:

[0006] A corporate credit assessment method based on semi-supervised learning, comprising:

[0007] Classify historical loan enterprises to obtain loan enterprises of several categories;

[0008] For each category of loan enterprises, select loan default enterprises and high-quality loan enterprises as representative enterprises;

[0009] Acquire and annotate enterprise data of loan default enterprises to obtain first annotated data, and construct a positive sample data set based on the first annotated data;

[0010] Acquire and annotate the enterprise data of high-quality loan enterprises to obtain second annotated data, and construct a negative sample data set based on the second annotated data;

[0011] Obtain default risk events, and generate several initial classifiers based on the positive sample data set, the negative sample data set and the default risk event training, where each category of loan enterprises corresponds to an initial classifier;

[0012] For each category of loan enterprises, obtain the enterprise data of loan enterprises other than the representative enterprise, and obtain unlabeled data of several loan enterprise categories;

[0013] For each initial classifier, iterative training is performed using labeled data and unlabeled data corresponding to the loan enterprise category to obtain several enterprise credit assessment classifiers, where each category of loan enterprise corresponds to an enterprise credit assessment classifier;

[0014] Combining several enterprise credit assessment classifiers to obtain an enterprise credit assessment model;

[0015] The enterprise data of the target loan enterprise is obtained, and the enterprise data of the target loan enterprise is input into the enterprise credit assessment model to obtain the credit assessment result of the target loan enterprise.

[0016] Furthermore, for each category of loan enterprises, the steps of selecting loan default enterprises and high-quality loan enterprises as representative enterprises specifically include:

[0017] Obtain default risk events, analyze them, and obtain default risk benchmarks;

[0018] For all loan enterprises in each category, use the default risk benchmark to conduct loan default risk assessment and obtain loan default risk assessment results;

[0019] According to the loan default risk assessment results, all loan enterprises in each category are ranked according to their loan default risk;

[0020] Select the enterprises with the highest loan default risk as loan default enterprises, and select the enterprises with the lowest loan default risk as high-quality loan enterprises;

[0021] Loan default enterprises and high-quality loan enterprises are taken as representative enterprises.

[0022] Furthermore, the enterprise data includes structured data and unstructured data. The steps of acquiring and labeling the enterprise data of loan default enterprises, obtaining first labeled data, and constructing a positive sample data set based on the first labeled data specifically include:

[0023] Acquire enterprise data of loan default enterprises from a preset database to obtain first enterprise data;

[0024] Classifying the first enterprise data to identify structured data and unstructured data in the first enterprise data;

[0025] Performing data structuring processing on unstructured data in the first enterprise data to obtain first transformed data;

[0026] Using the default risk benchmark, data annotating the structured data and the first transformed data in the first enterprise data to obtain first annotated data;

[0027] A positive sample dataset is constructed based on the first labeled data.

[0028] Further, the steps of acquiring and labeling the enterprise data of high-quality loan enterprises, obtaining second labeled data, and constructing a negative sample data set based on the second labeled data specifically include:

[0029] Acquire enterprise data of high-quality loan enterprises from a database to obtain second enterprise data;

[0030] Classifying the second enterprise data and identifying structured data and unstructured data in the second enterprise data;

[0031] Performing data structuring processing on the unstructured data in the second enterprise data to obtain second transformed data;

[0032] Using the default risk benchmark, data annotating is performed on the structured data and the second transformed data in the second enterprise data to obtain second annotated data;

[0033] A negative sample dataset is constructed based on the second labeled data.

[0034] Furthermore, the unstructured data includes text data, voice data, image data and video data. The step of performing data structuring processing on the unstructured data in the first enterprise data to obtain the first converted data specifically includes:

[0035] Performing text feature extraction on the text data in the first enterprise data to obtain first text data features;

[0036] Performing speech-to-text processing on the speech data in the first enterprise data, and performing text feature extraction on the obtained speech text to obtain a first speech-to-text feature;

[0037] Performing image feature extraction on the image data in the first enterprise data to obtain first image data features;

[0038] Extracting video key frames from the video data in the first enterprise data, and extracting image features from the extracted video key frames to obtain first video key frame features;

[0039] The first text data feature, the first speech text feature, the first image data feature and the first video key frame feature are combined to obtain first conversion data.

[0040] Furthermore, the step of performing data structuring processing on the unstructured data in the second enterprise data to obtain the second transformed data specifically includes:

[0041] Performing text feature extraction on the text data in the second enterprise data to obtain second text data features;

[0042] Performing speech-to-text processing on the speech data in the second enterprise data, and performing text feature extraction on the obtained speech text to obtain a second speech-to-text feature;

[0043] Performing image feature extraction on the image data in the second enterprise data to obtain second image data features;

[0044] Extracting video key frames from the video data in the second enterprise data, and extracting image features from the extracted video key frames to obtain second video key frame features;

[0045] The second text data feature, the second speech text feature, the second image data feature and the second video key frame feature are combined to obtain second conversion data.

[0046] Furthermore, based on the training of the gradient boosting decision tree ensemble algorithm, a corporate credit assessment classifier is generated. For each initial classifier, labeled data and unlabeled data of the corresponding loan enterprise category are used for iterative training to obtain several corporate credit assessment classifiers, specifically including:

[0047] For each initial classifier, randomly and with replacement sample the target unlabeled data from the unlabeled data of the corresponding loan enterprise category;

[0048] Construct a new data training set based on the positive sample data set, negative sample data set and target unlabeled data;

[0049] Use the initial classifier to perform data classification learning on the new data training set, and adjust the classifier parameters of the initial classifier according to the data classification learning results;

[0050] Each initial classifier is trained iteratively until every unlabeled data is involved in data classification learning, and several corporate credit assessment classifiers are obtained.

[0051] In order to solve the above technical problems, the embodiment of the present application also provides an enterprise credit assessment device based on semi-supervised learning, which adopts the following technical solution:

[0052] An enterprise credit assessment device based on semi-supervised learning, comprising:

[0053] The enterprise classification module is used to classify historical loan enterprises and obtain loan enterprises of several categories;

[0054] The enterprise representative module is used to select loan default enterprises and high-quality loan enterprises as representative enterprises for each category of loan enterprises;

[0055] A first labeling module is used to obtain and label the enterprise data of loan default enterprises, obtain first labeled data, and construct a positive sample data set based on the first labeled data;

[0056] The second labeling module is used to obtain and label the enterprise data of high-quality loan enterprises, obtain second labeled data, and construct a negative sample data set based on the second labeled data;

[0057] The first classification module is used to obtain default risk events and generate several initial classifiers based on the positive sample data set, the negative sample data set and the default risk event training, wherein each category of loan enterprises corresponds to an initial classifier;

[0058] The data acquisition module is used to acquire enterprise data of loan enterprises other than the representative enterprise for each category of loan enterprises, and obtain unlabeled data of several loan enterprise categories;

[0059] The second classification module is used to iteratively train each initial classifier using labeled data and unlabeled data corresponding to the loan enterprise category to obtain a number of enterprise credit assessment classifiers, wherein each category of loan enterprise corresponds to an enterprise credit assessment classifier;

[0060] A classifier combination module is used to combine several enterprise credit assessment classifiers to obtain an enterprise credit assessment model;

[0061] The credit assessment module is used to obtain the enterprise data of the target loan enterprise and input the enterprise data of the target loan enterprise into the enterprise credit assessment model to obtain the credit assessment result of the target loan enterprise.

[0062] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0063] A computer device comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the enterprise credit assessment method based on semi-supervised learning as described in any one of the above are implemented.

[0064] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0065] A computer-readable storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement the steps of the enterprise credit assessment method based on semi-supervised learning as described in any one of the above.

[0066] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0067] The present application provides a corporate credit assessment method based on semi-supervised learning and related equipment. The present application provides a corporate credit assessment solution based on semi-supervised learning. By utilizing a small amount of labeled data and a large amount of unlabeled data, the model's ability to assess corporate credit risk is significantly improved, while reducing the workload of data labeling and the impact of human factors. The present application constructs classifiers for different categories of corporate characteristics, which can capture the unique credit risk characteristics of each category and enhance the adaptability and accuracy of the model; through iterative training, the potential information of unlabeled data is fully mined, so that the model has higher generalization ability and robustness; at the same time, combined with the integration strategy of multiple classifiers, the stability and accuracy of the assessment results are further improved. The overall model can efficiently identify the credit level and default risk of the loan enterprise, provide scientific decision-making support for financial institutions, and help reduce credit risk and optimize resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0069] Figure 1 An exemplary system architecture diagram is shown in which the present application can be applied;

[0070] Figure 2 A flowchart of an embodiment of a method for evaluating corporate credit based on semi-supervised learning according to the present application is shown;

[0071] Figure 3A schematic diagram of the structure of an embodiment of an enterprise credit assessment device based on semi-supervised learning according to the present application is shown;

[0072] Figure 4 A schematic structural diagram of an embodiment of a computer device according to the present application is shown. DETAILED DESCRIPTION

[0073] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0074] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0075] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0076] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0077] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0078] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to a laptop computer 1011, a tablet computer 1012 or a mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0079] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0080] It should be noted that the enterprise credit assessment method based on semi-supervised learning provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the enterprise credit assessment device based on semi-supervised learning is generally set in the server / terminal device.

[0081] It should be understood that Figure 1 The numbers of terminal devices, networks and servers in the embodiment are merely illustrative. The above system may have any number of terminal devices, networks and servers according to implementation requirements.

[0082] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for corporate credit assessment based on semi-supervised learning according to the present application. The method for corporate credit assessment based on semi-supervised learning comprises the following steps:

[0083] S201, classify the historical loan enterprises to obtain loan enterprises of several categories.

[0084] Specifically, by analyzing the attributes of historical loan companies, we select distinguishing features such as industry type, company size, and regional distribution to form several categories. These categories reflect the characteristics and potential risks of different loan companies, ensuring that the assessment model can be well adapted to different types of companies.

[0085] S202, for each category of loan enterprises, select loan default enterprises and high-quality loan enterprises as representative enterprises.

[0086] Specifically, based on the loan history, enterprises with a clear default history and typical performance are selected from each category as default representatives, and enterprises with stable finances and good credit ratings are selected as high-quality representatives. The selection of representative enterprises needs to cover the main features within the category to ensure the representativeness of the labeled data and the discrimination of the samples. One loan default enterprise and one high-quality loan enterprise are selected as representative enterprises, or multiple loan default enterprises and multiple high-quality loan enterprises can be selected as representative enterprises.

[0087] For example, in each category of loan enterprises, according to the existing default risk events, the loan enterprises are divided into time periods, and the defaulting enterprises are selected as high-risk positive samples. These enterprises are enterprises that fail to fulfill their repayment obligations on time within a certain period of time. At the same time, high-quality loan enterprises are marked as low-risk negative samples. These enterprises can fulfill their debts on time and have not encountered any defaults or potential risks.

[0088] For another example, we select representative enterprises that have historical default risk events and high-quality loan enterprises judged by business personnel; in the labeling of positive and negative sample data of loan enterprises, since the credit of enterprises will change with the internal business conditions and macro policies of enterprises, it is necessary to divide the relevant characteristics of loan enterprises into time periods. By obtaining historical default risk events, the characteristics of loan enterprises that have defaulted in the predetermined time period are taken as positive samples, while the high-quality loan enterprises judged by business personnel are regarded as negative samples.

[0089] S203, obtaining and labeling enterprise data of loan defaulting enterprises to obtain first labeled data, and constructing a positive sample data set based on the first labeled data.

[0090] Specifically, we collect detailed corporate data representing defaulting companies, including financial data, operating data, and public opinion data, and confirm them as positive samples based on expert annotations and default records. By organizing and cleaning these data, we construct a positive sample data set with high-risk characteristics.

[0091] S204, obtaining and labeling the enterprise data of high-quality loan enterprises to obtain second labeled data, and constructing a negative sample data set based on the second labeled data.

[0092] Specifically, we extract key data of high-quality representative enterprises, such as good debt repayment records, stable revenue growth, low debt ratio, etc., and mark them as negative samples. By constructing a negative sample data set containing the characteristics of low-risk enterprises, we provide clear positive guidance for the model to learn the excellent characteristics of corporate credit.

[0093] S205, obtaining default risk events, and generating a number of initial classifiers based on the positive sample data set, the negative sample data set and the default risk event training, wherein each category of loan enterprises corresponds to an initial classifier.

[0094] Specifically, we collect data on external risk events that affect loan defaults (such as economic recessions and policy changes), combine them with labeled data, and use classification algorithms (such as logistic regression and random forests) to generate initial classifiers. Each classifier targets different categories of enterprises and captures the credit assessment characteristics and external risk impacts unique to that category.

[0095] In a specific embodiment of the present application, the initial classifier is built by referring to the Two-step PU Learning method. Two-step PU Learning is an implementation method of PU learning (positive and unlabeled learning). PU learning is a semi-supervised binary classification method that learns from positive samples (Positive) and unlabeled samples (Unlabeled). By learning from the positive samples in the data and applying the learned knowledge to relabel unknown samples, it is suitable for any scenario that requires binary classification of unreliable data. The basic idea of ​​Two-step PU Learning is to learn in two stages: first, use positive samples and unlabeled samples to train a preliminary classifier; then, use the classifier to predict the unlabeled samples, use the data predicted as negative samples as negative samples, and retrain a new classifier, which is suitable for scenarios where negative examples account for the majority of unlabeled samples. In the scenario of corporate credit assessment, the proportion of low-risk enterprises is larger, and highly reliable initial positive and negative samples are obtained based on certain rules. Therefore, this application adopts an improved method of Two-step PU Learning, using positive sample data sets, negative sample data sets and target unlabeled data randomly sampled from unlabeled data to construct an iterative training sample set for classifier training.

[0096] S206, for each category of loan enterprises, obtain the enterprise data of loan enterprises other than the representative enterprise, and obtain unlabeled data of several loan enterprise categories.

[0097] Specifically, all loan enterprise data except the representative enterprise are extracted from the category. These unlabeled data cover a wide range of enterprise characteristics and potential risk information. These data are not labeled as positive or negative samples and need to be effectively utilized through further training of the classifier to improve the generalization ability of the model.

[0098] S207, for each initial classifier, iterative training is performed using labeled data and unlabeled data corresponding to the loan enterprise category to obtain a plurality of enterprise credit assessment classifiers, wherein each category of loan enterprise corresponds to an enterprise credit assessment classifier.

[0099] Specifically, the initial classifier predicts the unlabeled data, relabels some of the unlabeled data based on the high confidence results, and adds the positive and negative sample data sets to iteratively train the classifier. After multiple rounds of iterations, the classifier can more accurately identify the features of the unlabeled data and gradually form a stable credit assessment classifier.

[0100] S208, combining several enterprise credit assessment classifiers to obtain an enterprise credit assessment model.

[0101] Specifically, multiple categories of credit assessment classifiers are combined into a unified model through integration methods such as weighted voting and average fusion. The integrated model fully utilizes the evaluation capabilities of different classifiers to ensure high accuracy and strong adaptability of data evaluation for cross-category enterprises, forming a universal credit assessment framework.

[0102] S209, obtaining the enterprise data of the target loan enterprise, and inputting the enterprise data of the target loan enterprise into the enterprise credit evaluation model to obtain the credit evaluation result of the target loan enterprise.

[0103] Specifically, relevant data of the target enterprise (such as financial status, repayment history, and industry performance) is collected and input into the enterprise credit assessment model. The model calculates the assessment results based on the input data and outputs the credit score or default probability, providing financial institutions with a clear decision-making basis to support loan approval and risk control.

[0104] In the above embodiment, this application draws on the Two-step PU Learning method to build an initial classifier, and through iterative training, fully explores the potential information of unlabeled data, so that the model has higher generalization ability and robustness; at the same time, combined with the integration strategy of multiple classifiers, the stability and accuracy of the evaluation results are further improved. The overall model can efficiently identify the credit level and default risk of the loan enterprise, provide scientific decision-making support for financial institutions, and help reduce credit risks and optimize resource allocation.

[0105] Furthermore, for each category of loan enterprises, the steps of selecting loan default enterprises and high-quality loan enterprises as representative enterprises specifically include:

[0106] Obtain default risk events, analyze them, and obtain default risk benchmarks;

[0107] For all loan enterprises in each category, use the default risk benchmark to conduct loan default risk assessment and obtain loan default risk assessment results;

[0108] According to the loan default risk assessment results, all loan enterprises in each category are ranked according to their loan default risk;

[0109] Select the enterprises with the highest loan default risk as loan default enterprises, and select the enterprises with the lowest loan default risk as high-quality loan enterprises;

[0110] Loan default enterprises and high-quality loan enterprises are taken as representative enterprises.

[0111] In this embodiment, obtaining default risk events refers to extracting default information and risk factors related to loan enterprises from historical data, such as economic recession, industry fluctuations, policy adjustments, etc., as default risk benchmarks. These default risk events are analyzed, and key characteristic indicators (such as industry impact coefficients, time nodes, etc.) are extracted to form a default risk benchmark with guiding significance for assessing the risk level of each enterprise. Next, for the loan enterprises in each category, the default risk benchmark is combined with the enterprise's financial, operational and other data, and its default risk value is calculated through the model to obtain the loan default risk assessment result. The assessment results of the enterprises in each category are sorted from the highest risk to the lowest. The top-ranked enterprise is selected as the loan default enterprise, indicating that the enterprise has representative high-risk characteristics; the last-ranked enterprise is selected as a high-quality loan enterprise, representing its low-risk characteristics. These two representative enterprises are used as typical samples of this category.

[0112] Through the above steps, high-risk and low-risk enterprises can be accurately identified within the category, ensuring the representativeness of the labeled samples, thereby improving the accuracy and adaptability of subsequent model evaluation.

[0113] Furthermore, the enterprise data includes structured data and unstructured data. The steps of acquiring and labeling the enterprise data of loan default enterprises, obtaining first labeled data, and constructing a positive sample data set based on the first labeled data specifically include:

[0114] Acquire enterprise data of loan default enterprises from a preset database to obtain first enterprise data;

[0115] Classifying the first enterprise data to identify structured data and unstructured data in the first enterprise data;

[0116] Performing data structuring processing on unstructured data in the first enterprise data to obtain first transformed data;

[0117] Using the default risk benchmark, data annotating the structured data and the first transformed data in the first enterprise data to obtain first annotated data;

[0118] A positive sample dataset is constructed based on the first labeled data.

[0119] Enterprise data includes financial data, operating data and public opinion data, all of which are multi-dimensional data. Financial data includes balance sheet, income statement, cash flow statement, etc., which can analyze changes in financial indicators such as profitability, debt repayment ability, and liquidity. Operating data includes sales revenue, equipment utilization, inventory turnover days, etc., reflecting changes in the business operations of the enterprise. Public opinion data includes news, Weibo public opinion, WeChat articles and their comments, judicial litigation data, and administrative penalty data of the target enterprise. It can analyze changes in negative public opinion of the enterprise, changes in the number of lawsuits as defendants, changes in the amount involved, and changes in the number of administrative penalties. At the same time, public opinion also includes industry policies and news. From the positive and negative emotions of industry news and the impact of policies or news on the stock price of industry sectors, we can analyze the impact of industry regulations on the future development of enterprises.

[0120] In this embodiment, the enterprise data of loan default enterprises is obtained from a preset database, which is the first enterprise data. Then, the first enterprise data is classified to distinguish between structured data (such as financial statements, credit scores, etc.) and unstructured data (such as contract texts, customer feedback, news reports, etc.). For unstructured data, natural language processing (NLP) technology or text mining methods are used to convert it into structured data, extract key features such as indicators related to enterprise risks, and generate first transformed data. Subsequently, the structured data and the first transformed data are annotated in combination with the default risk benchmark to clarify whether they reflect high-risk characteristics, thereby forming the first annotated data. Finally, the first annotated data is summarized based on the annotation results to construct a positive sample data set.

[0121] Through the above steps, we can effectively integrate and utilize structured and unstructured data, improve the accuracy of data annotation, and construct a representative positive sample data set.

[0122] Further, the steps of acquiring and labeling the enterprise data of high-quality loan enterprises, obtaining second labeled data, and constructing a negative sample data set based on the second labeled data specifically include:

[0123] Acquire enterprise data of high-quality loan enterprises from a database to obtain second enterprise data;

[0124] Classifying the second enterprise data and identifying structured data and unstructured data in the second enterprise data;

[0125] Performing data structuring processing on the unstructured data in the second enterprise data to obtain second transformed data;

[0126] Using the default risk benchmark, data annotating is performed on the structured data and the second transformed data in the second enterprise data to obtain second annotated data;

[0127] A negative sample dataset is constructed based on the second labeled data.

[0128] In this embodiment, the enterprise data of high-quality loan enterprises, i.e., the second enterprise data, is obtained from the database. Then, the second enterprise data is classified to identify the structured data (such as financial statements, loan records, credit ratings, etc.) and unstructured data (such as news reports, customer reviews, industry analysis, etc.). For unstructured data, natural language processing (NLP) or machine learning technology is used for structured processing, important features are extracted and converted into quantitative data to form the second transformed data. Then, combined with the default risk benchmark, the structured data and the second transformed data are annotated to ensure accurate identification of high-quality enterprise characteristics and generate the second annotated data. Finally, the second annotated data is sorted and summarized according to the annotation results to construct a negative sample data set.

[0129] Through the above steps, we can make full use of the data of high-quality loan companies, build a high-quality negative sample data set, and improve the model's recognition ability and classification effect of low-risk features.

[0130] Furthermore, the unstructured data includes text data, voice data, image data and video data. The step of performing data structuring processing on the unstructured data in the first enterprise data to obtain the first converted data specifically includes:

[0131] Performing text feature extraction on the text data in the first enterprise data to obtain first text data features;

[0132] Performing speech-to-text processing on the speech data in the first enterprise data, and performing text feature extraction on the obtained speech text to obtain a first speech-to-text feature;

[0133] Performing image feature extraction on the image data in the first enterprise data to obtain first image data features;

[0134] Extracting video key frames from the video data in the first enterprise data, and extracting image features from the extracted video key frames to obtain first video key frame features;

[0135] The first text data feature, the first speech text feature, the first image data feature and the first video key frame feature are combined to obtain first conversion data.

[0136] In this embodiment, the unstructured data includes text data, voice data, image data and video data. First, for text data, natural language processing (NLP) technology is used to extract text features, including keyword extraction, sentiment analysis and topic modeling, to generate a first text data feature reflecting the text content. Secondly, for voice data, voice recognition technology is used to convert voice into text, and the converted text data is subjected to feature extraction through NLP technology to obtain a first voice text feature. Then, for image data, image processing technology is used to extract features, such as identifying key objects, color distribution and texture features in the image, to generate a first image data feature. In addition, for video data, video key frame extraction is first performed to reduce redundant information and extract important content, and then the image feature extraction method is applied to the key frame image to obtain the first video key frame feature. Finally, the above four types of features (text, voice text, image, video key frame) are combined to form a structured first converted data.

[0137] Through the above steps, efficient structured transformation of unstructured data is achieved, the data value is maximized, multi-dimensional, high-quality input data is provided for the model, and the comprehensiveness and accuracy of the evaluation are improved.

[0138] Furthermore, the step of performing data structuring processing on the unstructured data in the second enterprise data to obtain the second transformed data specifically includes:

[0139] Performing text feature extraction on the text data in the second enterprise data to obtain second text data features;

[0140] Performing speech-to-text processing on the speech data in the second enterprise data, and performing text feature extraction on the obtained speech text to obtain a second speech-to-text feature;

[0141] Performing image feature extraction on the image data in the second enterprise data to obtain second image data features;

[0142] Extracting video key frames from the video data in the second enterprise data, and extracting image features from the extracted video key frames to obtain second video key frame features;

[0143] The second text data feature, the second speech text feature, the second image data feature and the second video key frame feature are combined to obtain second conversion data.

[0144] In this embodiment, first, for text data, text features are extracted by natural language processing (NLP) technology, such as word frequency statistics, sentiment analysis, and topic modeling, to generate second text data features that reflect the characteristics of high-quality enterprises. Secondly, for voice data, voice recognition technology is used to convert voice into text, and then important information in the voice text, such as positive sentiment words, service quality description, etc., is extracted by NLP method to generate second voice text features. Subsequently, for image data, image analysis technology is applied to extract features, such as color distribution, image texture, and object recognition, to form second image data features. For video data, the frame that best represents the video content is selected by key frame extraction technology, and image features are extracted from the key frame to generate the second video key frame features. Finally, the above features (text, voice text, image, video key frame) are organically integrated to generate the second conversion data.

[0145] Through the above steps, unstructured data can be processed comprehensively and efficiently, and the key features of high-quality loan companies can be fully extracted, providing a reliable and high-quality data foundation for the construction of negative sample data sets, thereby improving the model's assessment ability and accuracy for low-risk companies.

[0146] Furthermore, the steps of generating an enterprise credit assessment classifier based on the preset classification model training, iteratively training each initial classifier using the labeled data and unlabeled data of the corresponding loan enterprise category to obtain several enterprise credit assessment classifiers specifically include:

[0147] For each initial classifier, randomly and with replacement sample the target unlabeled data from the unlabeled data of the corresponding loan enterprise category as supplementary negative sample data;

[0148] Construct a new data training set based on the positive sample data set, negative sample data set and target unlabeled data;

[0149] Use the initial classifier to perform data classification learning on the new data training set, and adjust the classifier parameters of the initial classifier according to the data classification learning results;

[0150] Each initial classifier is trained iteratively until every unlabeled data is involved in data classification learning, and several corporate credit assessment classifiers are obtained.

[0151] The initial classifier is built based on the XGBoost classification model. XGBoost (eXtreme Gradient Boosting) is an ensemble learning algorithm based on the Gradient Boosting Decision Tree (GBDT). XGBoost builds an ensemble model by combining multiple weak learners (usually decision trees). Its core principle involves the optimization of loss functions and the construction of tree models. XGBoost is widely used to solve problems such as classification, regression, sorting, anomaly detection, and model interpretation. In classification problems, XGBoost can be used for binary and multi-classification tasks.

[0152] In this embodiment, in the process of training and generating a corporate credit assessment classifier based on the XGBoost integrated algorithm, first, for each initial classifier, it is necessary to sample the unlabeled data of the corresponding loan enterprise category as a supplementary negative sample and put it into the training set. This process is to extract the target data from the unlabeled data in a random and put-back manner. This method can ensure that the selected samples in each iteration are representative and reduce the deviation in the training process. Then, a new training set is constructed based on the positive sample data set, the negative sample data set and the target unlabeled data. This training set contains known positive and negative samples and the currently extracted unlabeled data, so as to ensure that the model can fully learn the characteristics of enterprises of different categories. Then, the initial classifier is used to perform data classification learning on the newly constructed training set, and the learning results are analyzed and the parameters of the initial classifier are adjusted according to the classification performance of the model. The adjustment process is usually to optimize the parameters of the classifier by the gradient descent method to improve the classification accuracy. The process is carried out through a loop iteration until all unlabeled data are involved in the training, thereby gradually improving the performance of the classifier. The obtained classifier is used to predict the unlabeled data that has not participated in the training, and samples with high confidence are selected to be added to the positive and negative sample data sets for subsequent iterative training. Finally, through this series of training steps, several independent corporate credit assessment classifiers can be obtained. These classifiers show good predictive ability on different categories of corporate data.

[0153] Through the above steps and the above iterative training process, the accuracy and generalization ability of each initial classifier can be gradually improved, ensuring that multiple corporate credit assessment classifiers can achieve good prediction results in different categories of loan companies.

[0154] In the above embodiments, the present application provides a corporate credit assessment method based on semi-supervised learning. The present application provides a corporate credit assessment scheme based on semi-supervised learning. By utilizing a small amount of labeled data and a large amount of unlabeled data, the model's ability to assess corporate credit risk is significantly improved, while reducing the workload of data labeling and the impact of human factors. The present application constructs classifiers for different categories of corporate characteristics, which can capture the unique credit risk characteristics of each category and enhance the adaptability and accuracy of the model; through iterative training, the potential information of unlabeled data is fully mined, so that the model has higher generalization ability and robustness; at the same time, combined with the integration strategy of multiple classifiers, the stability and accuracy of the assessment results are further improved. The overall model can efficiently identify the credit level and default risk of the loan enterprise, provide scientific decision-making support for financial institutions, and help reduce credit risks and optimize resource allocation.

[0155] In this embodiment, the enterprise credit assessment method based on semi-supervised learning is executed on the electronic device (eg Figure 1 The server shown in the figure) can receive instructions or obtain data through a wired connection or a wireless connection. It should be noted that the above wireless connection method may include but is not limited to 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0156] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0157] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0158] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0159] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0160] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of an enterprise credit assessment device based on semi-supervised learning. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0161] like Figure 3 As shown, the enterprise credit assessment device 300 based on semi-supervised learning described in this embodiment includes:

[0162] The enterprise classification module 301 is used to classify the historical loan enterprises to obtain loan enterprises of several categories;

[0163] The enterprise representative module 302 is used to select loan default enterprises and high-quality loan enterprises as representative enterprises for each category of loan enterprises;

[0164] The first labeling module 303 is used to obtain and label the enterprise data of loan default enterprises, obtain first labeled data, and construct a positive sample data set based on the first labeled data;

[0165] The second labeling module 304 is used to obtain and label the enterprise data of high-quality loan enterprises, obtain second labeled data, and construct a negative sample data set based on the second labeled data;

[0166] The first classification module 305 is used to obtain default risk events and generate a number of initial classifiers based on the positive sample data set, the negative sample data set and the default risk event training, wherein each category of loan enterprises corresponds to an initial classifier;

[0167] The data acquisition module 306 is used to acquire enterprise data of loan enterprises other than the representative enterprise for each category of loan enterprises, and obtain unlabeled data of several loan enterprise categories;

[0168] The second classification module 307 is used to iteratively train each initial classifier using the labeled data and unlabeled data corresponding to the loan enterprise category to obtain a plurality of enterprise credit assessment classifiers, wherein each category of loan enterprise corresponds to one enterprise credit assessment classifier;

[0169] A classifier combination module 308 is used to combine several enterprise credit assessment classifiers to obtain an enterprise credit assessment model;

[0170] The credit evaluation module 309 is used to obtain the enterprise data of the target loan enterprise and input the enterprise data of the target loan enterprise into the enterprise credit evaluation model to obtain the credit evaluation result of the target loan enterprise.

[0171] Furthermore, the enterprise representative module 302 specifically includes:

[0172] An event analysis unit is used to obtain default risk events and analyze the default risk events to obtain a default risk benchmark;

[0173] The risk assessment unit is used to conduct loan default risk assessment for all loan enterprises in each category using the default risk benchmark to obtain loan default risk assessment results;

[0174] A risk ranking unit is used to rank the loan default risks of all loan enterprises in each category according to the loan default risk assessment results;

[0175] An enterprise selection unit is used to select enterprises with the highest loan default risk as loan default enterprises, and to select enterprises with the lowest loan default risk as high-quality loan enterprises;

[0176] The enterprise representative unit is used to identify loan default enterprises and high-quality loan enterprises as representative enterprises.

[0177] Furthermore, the enterprise data includes structured data and unstructured data, and the first annotation module 303 specifically includes:

[0178] A first acquisition unit is used to acquire enterprise data of loan default enterprises from a preset database to obtain first enterprise data;

[0179] A first classification unit, used to classify the first enterprise data and identify structured data and unstructured data in the first enterprise data;

[0180] A first structured processing unit, configured to perform data structured processing on unstructured data in the first enterprise data to obtain first transformed data;

[0181] A first data labeling unit, configured to label the structured data and the first transformed data in the first enterprise data using a default risk benchmark to obtain first labeled data;

[0182] The positive sample data set construction unit is used to construct a positive sample data set based on the first labeled data.

[0183] Furthermore, the second marking module 304 specifically includes:

[0184] A second acquisition unit is used to acquire enterprise data of high-quality loan enterprises from a database to obtain second enterprise data;

[0185] A second classification unit, used to classify the second enterprise data and identify structured data and unstructured data in the second enterprise data;

[0186] A second structured processing unit, configured to perform data structured processing on the unstructured data in the second enterprise data to obtain second transformed data;

[0187] A second data annotation unit is used to annotate the structured data and the second transformed data in the second enterprise data using the default risk benchmark to obtain second annotated data;

[0188] The negative sample data set construction unit is used to construct a negative sample data set based on the second labeled data.

[0189] Furthermore, the unstructured data includes text data, voice data, image data and video data, and the first structured processing unit specifically includes:

[0190] A first text feature subunit is used to extract text features from text data in the first enterprise data to obtain first text data features;

[0191] A first voice feature subunit is used to perform voice-to-text processing on the voice data in the first enterprise data, and extract text features from the obtained voice text to obtain a first voice-to-text feature;

[0192] A first image feature subunit is used to extract image features from the image data in the first enterprise data to obtain first image data features;

[0193] A first video feature subunit is used to extract video key frames from the video data in the first enterprise data, and to extract image features from the extracted video key frames to obtain first video key frame features;

[0194] The first feature combination subunit is used to combine the first text data feature, the first voice text feature, the first image data feature and the first video key frame feature to obtain the first conversion data.

[0195] Furthermore, the step of performing data structuring processing on the unstructured data in the second enterprise data to obtain the second transformed data specifically includes:

[0196] A second text feature subunit is used to extract text features from the text data in the second enterprise data to obtain second text data features;

[0197] The second voice feature subunit is used to perform voice-to-text processing on the voice data in the second enterprise data, and perform text feature extraction on the obtained voice text to obtain a second voice-to-text feature;

[0198] A second image feature subunit is used to extract image features from the image data in the second enterprise data to obtain second image data features;

[0199] The second video feature subunit is used to extract video key frames from the video data in the second enterprise data, and extract image features from the extracted video key frames to obtain second video key frame features;

[0200] The second feature combination subunit is used to combine the second text data feature, the second voice text feature, the second image data feature and the second video key frame feature to obtain second conversion data.

[0201] Furthermore, based on the training of the gradient boosting decision tree ensemble algorithm to generate the enterprise credit assessment classifier, the second classification module 307 specifically includes:

[0202] A data extraction unit, for randomly and with replacement sampling of target unlabeled data from the unlabeled data of the corresponding loan enterprise category for each initial classifier;

[0203] A training set construction unit, used to construct a new data training set based on a positive sample data set, a negative sample data set and target unlabeled data;

[0204] A classification learning unit, used to perform data classification learning on a new data training set using an initial classifier, and to adjust classifier parameters of the initial classifier according to the data classification learning result;

[0205] The loop iteration unit is used to perform loop iteration training for each initial classifier until each unlabeled data participates in the data classification learning, thereby obtaining a plurality of enterprise credit assessment classifiers.

[0206] In the above embodiments, the present application provides a corporate credit assessment device based on semi-supervised learning. The present application provides a corporate credit assessment solution based on semi-supervised learning. By utilizing a small amount of labeled data and a large amount of unlabeled data, the model's ability to assess corporate credit risk is significantly improved, while reducing the workload of data labeling and the impact of human factors. The present application constructs classifiers for different categories of corporate characteristics, which can capture the unique credit risk characteristics of each category and enhance the adaptability and accuracy of the model; through iterative training, the potential information of unlabeled data is fully mined, so that the model has higher generalization ability and robustness; at the same time, combined with the integration strategy of multiple classifiers, the stability and accuracy of the assessment results are further improved. The overall model can efficiently identify the credit level and default risk of the loan enterprise, provide scientific decision-making support for financial institutions, and help reduce credit risks and optimize resource allocation.

[0207] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0208] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 4 with a memory 41, a processor 42, and a network interface 43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0209] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0210] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 4. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the enterprise credit assessment method based on semi-supervised learning, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0211] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run the computer-readable instructions stored in the memory 41 or process data, such as running the computer-readable instructions of the enterprise credit assessment method based on semi-supervised learning.

[0212] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0213] In the above embodiment, the present application provides a computer device. The present application is based on a corporate credit assessment solution of semi-supervised learning. By utilizing a small amount of labeled data and a large amount of unlabeled data, the model's ability to assess corporate credit risk is significantly improved, while reducing the workload of data labeling and the impact of human factors. The present application constructs classifiers for different categories of corporate characteristics, which can capture the unique credit risk characteristics of each category and enhance the adaptability and accuracy of the model; through iterative training, the potential information of unlabeled data is fully mined, so that the model has higher generalization ability and robustness; at the same time, combined with the integration strategy of multiple classifiers, the stability and accuracy of the assessment results are further improved. The overall model can efficiently identify the credit level and default risk of the loan enterprise, provide scientific decision-making support for financial institutions, and help reduce credit risks and optimize resource allocation.

[0214] The present application also provides another embodiment, namely, providing a computer-readable storage medium, which stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the corporate credit assessment method based on semi-supervised learning as described above.

[0215] In the above embodiments, the present application provides a computer-readable storage medium. The present application is based on a semi-supervised learning enterprise credit assessment solution. By utilizing a small amount of labeled data and a large amount of unlabeled data, the model's ability to assess enterprise credit risk is significantly improved, while reducing the workload of data labeling and the impact of human factors. The present application constructs classifiers for different categories of enterprise characteristics, which can capture the unique credit risk characteristics of each category and enhance the adaptability and accuracy of the model; through iterative training, the potential information of unlabeled data is fully mined, so that the model has higher generalization ability and robustness; at the same time, combined with the integration strategy of multiple classifiers, the stability and accuracy of the assessment results are further improved. The overall model can efficiently identify the credit level and default risk of the loan enterprise, provide scientific decision-making support for financial institutions, and help reduce credit risks and optimize resource allocation.

[0216] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0217] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0218] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A corporate credit assessment method based on semi-supervised learning, characterized in that: include: Classify historical loan enterprises to obtain several categories of loan enterprises; For each category of loan enterprises, select loan default enterprises and high-quality loan enterprises as representative enterprises; Acquire and annotate the enterprise data of the loan defaulting enterprise to obtain first annotated data, and construct a positive sample data set based on the first annotated data; Acquire and annotate the enterprise data of the high-quality loan enterprises to obtain second annotated data, and construct a negative sample data set based on the second annotated data; Acquire default risk events, and generate a plurality of initial classifiers based on the positive sample data set, the negative sample data set, and the default risk event training, wherein each category of loan enterprises corresponds to an initial classifier; For each category of loan enterprises, obtain the enterprise data of loan enterprises other than the representative enterprise to obtain unlabeled data of several loan enterprise categories; For each initial classifier, iterative training is performed using labeled data and unlabeled data corresponding to the loan enterprise category to obtain several enterprise credit assessment classifiers, where each category of loan enterprise corresponds to an enterprise credit assessment classifier; Combining a plurality of the enterprise credit assessment classifiers to obtain an enterprise credit assessment model; The enterprise data of the target loan enterprise is obtained, and the enterprise data of the target loan enterprise is input into the enterprise credit evaluation model to obtain the credit evaluation result of the target loan enterprise.

2. The enterprise credit assessment method based on semi-supervised learning as claimed in claim 1, characterized in that: The steps of selecting loan default enterprises and high-quality loan enterprises as representative enterprises for each category of loan enterprises specifically include: Obtaining the default risk event, and analyzing the default risk event to obtain a default risk benchmark; For all loan enterprises in each category, use the default risk benchmark to conduct loan default risk assessment to obtain loan default risk assessment results; Rank all loan enterprises in each category by loan default risk according to the loan default risk assessment results; Selecting the enterprise with the highest loan default risk as the loan default enterprise, and selecting the enterprise with the lowest loan default risk as the high-quality loan enterprise; The loan default enterprises and the high-quality loan enterprises are regarded as the representative enterprises.

3. The enterprise credit assessment method based on semi-supervised learning as claimed in claim 2, characterized in that: The enterprise data includes structured data and unstructured data. The steps of acquiring and labeling the enterprise data of the loan default enterprise to obtain first labeled data and constructing a positive sample data set based on the first labeled data specifically include: Acquire the enterprise data of the loan default enterprise from a preset database to obtain first enterprise data; Classifying the first enterprise data to identify structured data and unstructured data in the first enterprise data; Performing data structuring processing on unstructured data in the first enterprise data to obtain first transformed data; Using the default risk benchmark, data annotating the structured data in the first enterprise data and the first transformed data to obtain the first annotated data; A positive sample data set is constructed based on the first labeled data.

4. The enterprise credit assessment method based on semi-supervised learning as claimed in claim 3, characterized in that: The step of acquiring and labeling the enterprise data of the high-quality loan enterprises to obtain second labeled data, and constructing a negative sample data set based on the second labeled data specifically includes: Acquire the enterprise data of the high-quality loan enterprise from the database to obtain second enterprise data; classifying the second enterprise data to identify structured data and unstructured data in the second enterprise data; Performing data structuring processing on the unstructured data in the second enterprise data to obtain second transformed data; Using the default risk benchmark, data annotating the structured data and the second transformed data in the second enterprise data to obtain the second annotated data; A negative sample dataset is constructed based on the second labeled data.

5. The enterprise credit assessment method based on semi-supervised learning as claimed in claim 4, characterized in that: The unstructured data includes text data, voice data, image data and video data. The step of performing data structuring processing on the unstructured data in the first enterprise data to obtain first converted data specifically includes: Performing text feature extraction on the text data in the first enterprise data to obtain first text data features; Performing speech-to-text processing on the speech data in the first enterprise data, and performing text feature extraction on the obtained speech text to obtain a first speech-to-text feature; Extracting image features from the image data in the first enterprise data to obtain first image data features; Extracting video key frames from the video data in the first enterprise data, and extracting image features from the extracted video key frames to obtain first video key frame features; The first text data feature, the first speech text feature, the first image data feature and the first video key frame feature are combined to obtain the first conversion data.

6. The enterprise credit assessment method based on semi-supervised learning as claimed in claim 5, characterized in that: The step of performing data structuring processing on the unstructured data in the second enterprise data to obtain the second transformed data specifically includes: Performing text feature extraction on the text data in the second enterprise data to obtain second text data features; Performing speech-to-text processing on the speech data in the second enterprise data, and performing text feature extraction on the obtained speech text to obtain a second speech-to-text feature; Performing image feature extraction on the image data in the second enterprise data to obtain second image data features; Extracting video key frames from the video data in the second enterprise data, and extracting image features from the extracted video key frames to obtain second video key frame features; The second text data feature, the second speech text feature, the second image data feature and the second video key frame feature are combined to obtain the second conversion data.

7. The enterprise credit assessment method based on semi-supervised learning as claimed in claim 1, characterized in that: The enterprise credit assessment classifier is generated based on the training of the gradient boosting decision tree ensemble algorithm, and the steps of iteratively training each initial classifier using the labeled data and unlabeled data of the corresponding loan enterprise category to obtain several enterprise credit assessment classifiers specifically include: For each initial classifier, randomly and with replacement sample the target unlabeled data from the unlabeled data of the corresponding loan enterprise category; Constructing a new data training set based on the positive sample data set, the negative sample data set and the target unlabeled data; Using the initial classifier to perform data classification learning on the new data training set, and adjusting classifier parameters of the initial classifier according to the data classification learning result; A cyclic iterative training is performed for each of the initial classifiers until each of the unlabeled data participates in the data classification learning, thereby obtaining a plurality of corporate credit assessment classifiers.

8. An enterprise credit assessment device based on semi-supervised learning, characterized in that: include: The enterprise classification module is used to classify historical loan enterprises and obtain loan enterprises of several categories; The enterprise representative module is used to select loan default enterprises and high-quality loan enterprises as representative enterprises for each category of loan enterprises; A first labeling module is used to obtain and label the enterprise data of the loan default enterprise to obtain first labeled data, and to construct a positive sample data set based on the first labeled data; A second labeling module is used to obtain and label the enterprise data of the high-quality loan enterprises to obtain second labeled data, and construct a negative sample data set based on the second labeled data; A first classification module is used to obtain default risk events and generate a plurality of initial classifiers based on the positive sample data set, the negative sample data set and the default risk event training, wherein each category of loan enterprises corresponds to an initial classifier; A data acquisition module is used to acquire enterprise data of loan enterprises other than the representative enterprise for each category of loan enterprises, and obtain unlabeled data of several loan enterprise categories; The second classification module is used to iteratively train each initial classifier using labeled data and unlabeled data corresponding to the loan enterprise category to obtain a number of enterprise credit assessment classifiers, wherein each category of loan enterprise corresponds to an enterprise credit assessment classifier; A classifier combination module, used to combine a plurality of the enterprise credit assessment classifiers to obtain an enterprise credit assessment model; The credit evaluation module is used to obtain the enterprise data of the target loan enterprise and input the enterprise data of the target loan enterprise into the enterprise credit evaluation model to obtain the credit evaluation result of the target loan enterprise.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the enterprise credit assessment method based on semi-supervised learning as described in any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the enterprise credit assessment method based on semi-supervised learning as described in any one of claims 1 to 7.

Images