CN-122023040-A - Enterprise credit automatic rating method integrating financial factors and non-financial factors

CN122023040ACN 122023040 ACN122023040 ACN 122023040ACN-122023040-A

Abstract

The invention relates to the technical field of financial data risk rating, in particular to an enterprise credit automatic rating method integrating financial and non-financial factors, which comprises the steps of constructing original financial data, analyzing the original financial data, constructing a credit characteristic index set comprising a financial characteristic index and a non-financial characteristic index, determining each characteristic layer and the importance weight corresponding to each characteristic layer based on the credit characteristic index set, and then feature index values of each feature layer and the importance weights corresponding to the feature index values are marked, a training set is generated and input into a pre-constructed deep learning model for training, and the enterprise credit grade is automatically output, so that the deep fusion of financial factors and non-financial factors is realized, and the accuracy, the instantaneity and the automation level of the enterprise credit grade are improved.

Inventors

LIU YANG
SONG ZIYU
GUO CHUNYANG

Assignees

全联征信有限公司

Dates

Publication Date: 20260512
Application Date: 20260128

Claims (9)

1. An enterprise credit automation rating method integrating financial and non-financial factors is characterized by comprising the following steps: collecting financial data and non-financial data corresponding to enterprises in a preset time window, and constructing original financial data; Analyzing the original financial data to construct a credit feature index set comprising financial feature indexes and non-financial feature indexes, Determining each feature layer and the importance weight corresponding to each feature layer based on the credit feature index set; the solving process of the importance weight is that in a preset time window, time sequence analysis is carried out on each characteristic index corresponding to each acquisition moment in each characteristic layer to obtain a time sequence state vector; Normalizing and weighting average is carried out on time sequence state vectors of all feature indexes in all feature layers to obtain a comprehensive state characterization value, and importance weights are determined according to preset reference intervals in which the comprehensive state characterization value falls; And (3) marking the characteristic index values of each characteristic layer and the corresponding importance weights thereof, generating a training set, inputting the training set into a pre-constructed deep learning model for training, and automatically outputting the credit rating of the enterprise.
2. The method for automatically rating credit of enterprises integrating financial factors and non-financial factors according to claim 1, wherein the preset time window is a time interval traced back forward based on the current rating time, and can be set as a year, a quarter, a month or a week.
3. The automated enterprise credit rating method integrating financial and non-financial factors as recited in claim 1, wherein the construction process of the financial characteristic index is as follows: extracting core financial fields corresponding to enterprises in a preset time window from the original financial data to construct a financial field set; performing field analysis and caliber unified processing on the financial field set by adopting a random forest field identification model, and performing time alignment processing on core financial data according to a preset time window to form a standardized financial data sequence with time continuity; and (3) carrying out index calculation on the financial condition of the enterprise according to a preset financial analysis rule based on the standardized financial data sequence, and constructing a financial characteristic index.
4. An automated corporate credit rating method combining financial and non-financial factors as recited in claim 3, wherein the financial characteristic indicators include flow rate, snap rate, equity rate, net sales rate, gross profit rate, operating profit rate, period rate, business cash flow rate, free cash flow, and cash flow structure stability rate.
5. The automated enterprise credit rating method for merging financial and non-financial factors as recited in claim 1, wherein the construction process of the non-financial characteristic index is as follows: Extracting core non-financial fields corresponding to enterprises in a preset time window from the original financial data to construct a non-financial field set; performing word segmentation, word stopping removal, word shape restoration and named entity recognition on the fields of each text in the non-financial field set to obtain a word sequence, and analyzing the word sequence to obtain negative topic intensity and negative emotion intensity; The negative topic intensity and the negative emotion intensity of each text form a non-financial characteristic index.
6. The automated enterprise credit rating method for merging financial and non-financial factors as recited in claim 5, wherein the process of solving for negative topic intensities is: inputting the word sequence into a topic analysis classifier constructed based on a pre-training language model, and judging topic categories of the terms in the word sequence by the topic analysis classifier to obtain topic labels corresponding to the terms in the word sequence, wherein the topic labels comprise a risk topic, a neutral topic and an opportunity topic; and respectively counting the lengths of different topic labels in the word sequence to obtain the term lengths corresponding to the risk topic, the neutral topic and the opportunity topic, and calculating the negative topic strength.
7. The automated corporate credit rating method combining financial and non-financial factors according to claim 5, wherein the process of solving for negative emotional intensity is: Inputting the word sequence into an emotion analysis classifier constructed based on a pre-training language model, judging emotion polarity of each term in the word sequence by the emotion analysis classifier, dividing the term into three types of positive words, neutral words or negative words, and respectively counting the number of corresponding terms to obtain the number of positive emotion words, the number of neutral emotion words and the number of negative emotion words; Adding the number of positive emotion words, the number of neutral emotion words and the number of negative emotion words to obtain the total number of emotion words; And then respectively carrying out duty ratio calculation on the number of positive emotion words, the number of neutral emotion words and the number of negative emotion words and the total number of emotion words to obtain positive emotion probability, neutral emotion probability and negative emotion probability, thereby calculating and obtaining negative emotion intensity.
8. The automated enterprise credit rating method for merging financial and non-financial factors as recited in claim 1, wherein each feature layer is determined based on a feature source, each feature layer comprising a financial layer, an operations management layer, an external risk layer, and an ESG layer.
9. The automated enterprise credit rating method for merging financial and non-financial factors as recited in claim 1, wherein the solving process of the time series state vector is as follows: Within a preset time window, set The first of the feature layers The time sequence formed by the characteristic indexes at each acquisition time is as follows: , wherein, The number of the feature layer is indicated, Representing the first in the feature layer The characteristic index of the characteristic is that, For a preset time window The acquisition time points are arranged in time sequence, Collecting the total number of moments in a time window; for any characteristic index, calculating the change rate between adjacent acquisition moments : ; On the basis, the average change rate of the characteristic index is obtained by carrying out average aggregation on the change rate in the whole time window: ; based on a time sequence formed by the characteristic indexes in a preset time window, carrying out statistical analysis on the time sequence, calculating the mean value of the characteristic indexes in the time dimension, further calculating the standard deviation of the mean value, and taking the standard deviation as the fluctuation amplitude for representing the characteristic indexes; Establishing a linear regression model between each acquisition time and a corresponding characteristic value, and obtaining a trend slope through least square fitting so as to determine a trend direction; and combining the average change rate, the fluctuation amplitude and the trend direction to construct a time sequence state vector of the characteristic index in a preset time window.

Description

Enterprise credit automatic rating method integrating financial factors and non-financial factors Technical Field The invention relates to the technical field of financial data risk rating, in particular to an enterprise credit automatic rating method integrating financial and non-financial factors. Background Enterprise credit rating is an important basis for financial risk management and credit decision-making, and accuracy and instantaneity of the enterprise credit rating directly influence risk control capability and market resource allocation efficiency of a financial institution; However, the conventional enterprise credit rating method mainly depends on financial statement analysis or expert experience judgment, and has various disadvantages: Firstly, the information source is single, and non-financial factors such as legal litigation, administrative punishment, media public opinion, management layer change and ESG information are ignored only by relying on financial data, so that enterprise risk identification has a blind area, secondly, the rating process is highly dependent on manual judgment, the influence of subjective experience is large, large-scale real-time updating is difficult to realize, and finally, the financial and non-financial information is not fused sufficiently, and the deep mining and comprehensive analysis of the nonlinear association relationship between the financial and non-financial information are not available, so that the comprehensiveness and the accuracy of credit rating are limited. In order to solve the above-mentioned defect, a technical scheme is provided. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides an enterprise credit automatic rating method integrating financial and non-financial factors, which can effectively solve the problems of single source of enterprise credit rating information, large dependence of manual intervention and insufficient integration of financial and non-financial information in the prior art. In order to achieve the above object, the present invention can be achieved by the following technical scheme: the invention provides an enterprise credit automatic rating method integrating financial and non-financial factors, which comprises the following steps: collecting financial data and non-financial data corresponding to enterprises in a preset time window, and constructing original financial data; Analyzing the original financial data to construct a credit feature index set comprising financial feature indexes and non-financial feature indexes, Determining each feature layer and the importance weight corresponding to each feature layer based on the credit feature index set; the solving process of the importance weight is that in a preset time window, time sequence analysis is carried out on each characteristic index corresponding to each acquisition moment in each characteristic layer to obtain a time sequence state vector; Normalizing and weighting average is carried out on time sequence state vectors of all feature indexes in all feature layers to obtain a comprehensive state characterization value, and importance weights are determined according to preset reference intervals in which the comprehensive state characterization value falls; And (3) marking the characteristic index values of each characteristic layer and the corresponding importance weights thereof, generating a training set, inputting the training set into a pre-constructed deep learning model for training, and automatically outputting the credit rating of the enterprise. Further, the preset time window is a time interval for backtracking forward based on the current rating time, and may be set to be annual, quaternary, monthly or weekly. Further, the construction process of the financial characteristic index comprises the following steps: extracting core financial fields corresponding to enterprises in a preset time window from the original financial data to construct a financial field set; performing field analysis and caliber unified processing on the financial field set by adopting a random forest field identification model, and performing time alignment processing on core financial data according to a preset time window to form a standardized financial data sequence with time continuity; and (3) carrying out index calculation on the financial condition of the enterprise according to a preset financial analysis rule based on the standardized financial data sequence, and constructing a financial characteristic index. Further, the financial characteristic indicators include flow rate, snap rate, equity rate, net sales rate, gross rate, operating profit rate, period rate, business cash flow rate, free cash flow, and cash flow structure stability rate. Further, the construction process of the non-financial characteristic index is as follows: Extracting core non-financial fields corresponding to enterprises in a preset time windo