Text-based banking business knowledge base processing device and method

By using a text-based banking business knowledge base processing device, the correlation between bank asset and liability business texts is calculated using cosine similarity and Jaccard similarity, which solves the shortcomings of interbank business correlation and risk assessment, and achieves more efficient and accurate risk prediction and monitoring.

CN117934140BActive Publication Date: 2024-11-08廊坊银行股份有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410136531.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-31
Publication Date
2024-11-08
Estimated Expiration
2044-01-31

AI Technical Summary

Technical Problem

Existing technologies lack dynamic analysis capabilities when assessing the business connections and systemic risks among commercial banks, and cannot effectively reflect the relationships between financial institutions. In particular, they are insufficient for risk assessment of small and medium-sized banks, and the existing methods have limited applicability, resulting in inaccurate risk characterization.

Method used

By using a text-based banking knowledge base processing device, the correlation between bank asset and liability business texts is calculated using cosine similarity and Jaccard similarity. Combined with machine learning and natural language processing technologies, the systemic risk level is predicted.

Benefits of technology

It improves the efficiency and accuracy of banking business correlation and risk prediction, enabling a more precise understanding of banking business development trends and prevention of potential risks, and is applicable to actual banking business operations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117934140B_ABST
    Figure CN117934140B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of banking business technology, and in particular to a text-based banking business knowledge base processing device and method. It includes: an acquisition unit, used to acquire the bank annual reports of at least two banks, and extract the corresponding asset business texts and liability business texts respectively; a conversion unit, used to convert and obtain a first word vector, a second word vector, a first text data word list and a second text data word list; a first calculation unit, used to calculate the correlation between the first word vector and the second word vector; a second calculation unit, used to calculate the correlation between the first text data word list and the second text data word list; a processing unit, used to determine the asset business correlation level between the two banks; a prediction unit, used to predict the systemic risk level according to the asset business correlation level. The present invention provides accurate support for predicting the systemic risk of commercial banks by introducing text data as an evaluation indicator of bank correlation.
Need to check novelty before this filing date? Find Prior Art

Description

Background Art

[0002] In the existing technology, the methods for measuring the business correlation and systemic risk of commercial banks are all based on market model method or indicator method. However, most of the indicator methods are based on current regulatory indicators, which are relatively simple to obtain data and can be combined with qualitative judgments. However, they are based on static markets and lack dynamic analysis. The market model method provides considerations in the time dimension based on high-frequency data to improve the forward-looking nature of risk measurement, but it has high requirements for the effectiveness of public market data. The existing financial market data still has shortcomings such as low data quality and small data samples. Neither the indicator method nor the market model method reflects the business connection relationship between financial institutions, lacks consideration of the correlation between financial institutions, and in recent years, the number of non-listed banks, especially small and medium-sized banks, has increased. In addition, they are small in scale and relatively fragile, and the hidden dangers of triggering systemic risks and regional risks cannot be ignored. The homogeneous operation characteristics of commercial banks are becoming increasingly prominent, and the impact of indirect correlation on the occurrence of systemic risks will also increase. The current methods based only on financial data are limited in their applicable objects, resulting in less description of the indirect correlation of asset-liability businesses and low adaptability to the current market. Therefore, how to provide a text-based banking business knowledge base processing device and method is a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0003] The purpose of the present invention is to provide a text-based banking business knowledge base processing device and method. By introducing text data as an evaluation indicator of bank association, the present invention broadens the data perspective of commercial bank business association measurement and systemically important bank identification research, and provides accurate support for predicting commercial bank systemic risks.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] A text-based banking business knowledge base processing device, comprising:

[0006] The acquisition unit is used to acquire the bank annual reports of at least two banks, and extract the asset business texts and liability business texts corresponding to the two banks respectively according to the two bank annual reports; wherein,

[0007] The asset business text includes a first asset business text and a second asset business text, and the liability business text includes a first liability business text and a second liability business text, wherein the first asset business text and the first liability business text correspond to one of the two banks, and the second asset business text and the second liability business text correspond to the other of the two banks;

[0008] a conversion unit, configured to convert the first asset business text and the first liability business text into a first word vector, and to convert the second asset business text and the second liability business text into a second word vector, the conversion unit further configured to convert the first asset business text and the first liability business text into a first text data word list, and to convert the second asset business text and the second liability business text into a second text data word list;

[0009] A first calculation unit, configured to calculate the correlation between the first word vector and the second word vector by using cosine similarity;

[0010] A second calculation unit, configured to calculate the association between the first text data word list and the second text data word list by using Jaccard similarity;

[0011] A processing unit, used for determining the asset business association level between the two banks according to the calculation results of the cosine similarity and the calculation results of the Jaccard similarity;

[0012] A prediction unit is used to predict the systemic risk level according to the asset-business association level.

[0013] In some embodiments of the present application, a text acquisition module is provided in the prediction unit, and the text acquisition module is used to acquire litigation data according to big data technology, and the litigation data includes judgment document network data and litigation case data;

[0014] The prediction unit is also provided with a text extraction module, which is further used to extract keywords from the litigation data and predict the credit risk level according to a preset risk identification model. The risk identification model has a number of preset keywords and a number of preset credit risk levels. The preset credit risk levels are divided from high to low into: a third-level credit risk level, a second-level credit risk level, and a first-level credit risk level; wherein,

[0015] When the consistency rate between the keywords appearing in the litigation data and the preset keywords in the risk identification model is greater than or equal to 30%, the first-level credit risk level is output; when the consistency rate between the preset keywords in the risk identification model is greater than or equal to 60%, the second-level credit risk level is output; when the consistency rate between the preset keywords in the risk identification model is greater than or equal to 90%, the third-level credit risk level is output;

[0016] The prediction unit is also provided with a text fine-tuning module, which is used to extract the keywords appearing in the litigation data when the consistency rate between the keywords appearing in the litigation data and several keywords preset in the risk identification model is less than 30%, and add several preset keywords that do not exist in the risk identification model to the risk identification model according to the extraction results.

[0017] In some embodiments of the present application, the first calculation unit is used to calculate the association between the first word vector and the second word vector by the following formula:

[0018]

[0019] Where, s im-cos x,y is the correlation between the first word vector and the second word vector, which is used to determine the asset business correlation between the two banks. i is the first word vector, y i is the second word vector.

[0020] In some embodiments of the present application, the processing unit is pre-set with a first preset cosine similarity correlation degree T01, a second preset cosine similarity correlation degree T02, a third preset cosine similarity correlation degree T03, and a fourth preset cosine similarity correlation degree T04, and 0<T01<T02<T03<T04<1;

[0021] The processing unit is pre-set with a first preset asset-business correlation level A1, a second preset asset-business correlation level A2, a third preset asset-business correlation level A3, and a fourth preset asset-business correlation level A4, and A1<A2<A3<A4;

[0022] The processing unit is further configured to select a corresponding asset-business association level as the asset-business association level between the two banks according to the relationship between the association calculation result j between the first word vector and the second word vector and each preset cosine similarity association level;

[0023] When j<T01, the first preset asset business association level A1 is selected as the asset business association level between the two banks;

[0024] When T01≤j<T02, the second preset asset business association level A2 is selected as the asset business association level between the two banks;

[0025] When T02≤j<T03, the third preset asset business association level A3 is selected as the asset business association level between the two banks;

[0026] When T03≤j<T04, the fourth preset asset-business association level A4 is selected as the asset-business association level between the two banks.

[0027] In some embodiments of the present application, the second calculation unit is used to calculate the association between the first text data word list and the second text data word list by the following formula:

[0028]

[0029] In the formula, s im-jaccard is the association between the first text data word list and the second text data word list, A is the set of the first text data word list, B is the set of the second text data word list, wherein the numerator represents the number of non-zero elements in the intersection of two vectors, and the denominator represents the number of non-zero elements in the union of two vectors.

[0030] In some embodiments of the present application, the processing unit is pre-set with a first preset Jaccard similarity correlation K01, a second preset Jaccard similarity correlation K02, and a third preset Jaccard similarity correlation K03, and K01<K02<K03;

[0031] The processing unit is further configured to increase each preset asset-business association level according to the relationship between the association calculation result i between the first text data word list and the second text data word list and each preset Jaccard similarity association level, and use the increased asset-business association level as the asset-business association level between the two banks;

[0032] When i<K01, and j<T01, the first preset asset business association level is raised by one level, and the second preset asset business association level is used as the asset business association level between the two banks;

[0033] When i<K01, and T01≤j<T02, the second preset asset business association level is raised by one level, and the third preset asset business association level is used as the asset business association level between the two banks;

[0034] When i<K01, and T02≤j<T03, the third preset asset business association level is raised by one level, and the fourth preset asset business association level is used as the asset business association level between the two banks;

[0035] When K01≤i<K02, and j<T01, the first preset asset business association level is raised by two levels, and the third preset asset business association level is used as the asset business association level between the two banks;

[0036] When K01≤i<K02, and T01≤j<T02, the second preset asset business association level is raised by two levels, and the fourth preset asset business association level is used as the asset business association level between the two banks;

[0037] When K02≤i<K03, and j<T01, the first preset asset business association level is upgraded by three levels, and the fourth preset asset business association level is used as the asset business association level between the two banks.

[0038] In some embodiments of the present application, the prediction unit is further used to calculate the systemic risk by the following formula:

[0039] mes i,t =α0+α1asset_cos i,t +α2lnscale i,t +α3fdirri i,t +α4epu i,t +α5npl i,t +α6ldr i,t +u i,t ;

[0040] In the formula, mes i,t is a systematic risk measurement indicator, asset-cos i,t is the asset business correlation between the two banks, lnscale i,t is the logarithm of total bank assets, epu i,t is the economic policy uncertainty index, fdirri i,t The foreign direct investment regulatory restriction index, npl i,t is the bank's non-performing loan ratio, idr i,t is the bank loan-to-deposit ratio, u i,t is the random disturbance term, a0 is the constant term, and a1 to a6 are regression coefficients.

[0041] In some embodiments of the present application, the prediction unit is further used to predict the systemic risk level according to the relationship between the calculated value g of the systemic risk and the preset systemic risk hazard value g0; wherein,

[0042] When g>g0, the first systemic risk level is used as the predicted systemic risk level;

[0043] When g>1.2g0, the second systemic risk level is used as the predicted systemic risk level;

[0044] When g>1.4g0, the third systemic risk level is used as the predicted systemic risk level;

[0045] When g>1.6g0, the fourth systemic risk level is taken as the predicted systemic risk level.

[0046] In some embodiments of the present application, the prediction unit is further set with a first preset risk compensation fund L01, a second preset risk compensation fund L02, a third preset risk compensation fund L03, and a fourth preset risk compensation fund L04, and L01<L02<L03<L04;

[0047] The prediction unit is also used to determine the generated risk compensation fund according to the predicted systemic risk level; wherein,

[0048] When the first systemic risk level is used as the predicted systemic risk level, determining the first preset risk compensation fund L01 as the generated risk compensation fund;

[0049] When the second systemic risk level is used as the predicted systemic risk level, determining the second preset risk compensation fund L02 as the generated risk compensation fund;

[0050] When the third systemic risk level is used as the predicted systemic risk level, determining the third preset risk compensation fund L03 as the generated risk compensation fund;

[0051] When the fourth systemic risk level is used as the predicted systemic risk level, the fourth preset risk compensation fund L04 is determined as the generated risk compensation fund.

[0052] In order to achieve the above-mentioned object, the present invention also provides a text-based banking business knowledge base processing method, which is applied to the text-based banking business knowledge base processing device, comprising:

[0053] Obtain bank annual reports of at least two banks, and extract asset business texts and liability business texts corresponding to the two banks respectively according to the two bank annual reports; wherein,

[0054] The asset business text includes a first asset business text and a second asset business text, and the liability business text includes a first liability business text and a second liability business text, wherein the first asset business text and the first liability business text correspond to one of the two banks, and the second asset business text and the second liability business text correspond to the other of the two banks;

[0055] According to the first asset business text and the first liability business text, the first word vector is converted, and according to the second asset business text and the second liability business text, the second word vector is converted, and the conversion unit is further used to convert the first asset business text and the first liability business text into a first text data word list, and according to the second asset business text and the second liability business text into a second text data word list;

[0056] Calculate the correlation between the first word vector and the second word vector by cosine similarity;

[0057] Calculating the association between the first text data word list and the second text data word list by Jaccard similarity;

[0058] Determine the asset business correlation level between the two banks according to the calculation results of the cosine similarity and the calculation results of the Jaccard similarity;

[0059] The systemic risk level is predicted based on the asset-business correlation level.

[0060] The present invention provides a text-based banking business knowledge base processing device and method, which has the following beneficial effects compared with the prior art:

[0061] The present invention processes and analyzes bank annual reports, extracts and compares asset business texts and liability business texts of different banks, thereby determining the degree of business correlation between banks and further predicting the systemic risk level. The invention combines automated processing and analysis to greatly improve the efficiency of business correlation and risk prediction. By using a large text model, the business texts of banks can be more accurately extracted and compared, thereby obtaining more accurate correlation and risk levels. In addition, by using machine learning and natural language processing technologies, the judgment process of business correlation and risk level is made more intelligent. The present invention is applied to actual banking business operations, and can effectively help banks and management departments better understand and monitor the development trend of banking business, as well as discover and prevent potential risks in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 is a functional block diagram of a text-based banking business knowledge base processing device in an embodiment of the present invention;

[0063] Figure 2 It is a flow chart of a text-based banking business knowledge base processing method in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0065] In the description of the present application, it should be understood that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0066] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0067] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a communication between the inner sides of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0068] In recent years, the number of non-listed banks, especially small and medium-sized banks, has increased. In addition, due to their small size and relative fragility, the hidden dangers of systemic and regional risks cannot be ignored. The homogeneous operation characteristics of commercial banks are becoming increasingly prominent, and the impact of indirect correlation on the occurrence of systemic risks will also increase. The current method based only on financial data is limited in its applicable objects, resulting in less description of indirect correlations between asset-liability businesses and problems such as low adaptability to the current market.

[0069] The purpose of the present invention is to provide a text-based banking business knowledge base processing device and method, which processes text data using natural language processing technology, adopts two text similarity indicators, cosine similarity and Jaccard similarity, as the basis for measuring the correlation between commercial banks, so as to characterize the indirect correlation brought about by the homogeneous operation between commercial banks, and finally form the correlation between the assets and liabilities of a single commercial bank and the overall correlation between the assets and liabilities of the commercial bank system.

[0070] See also Figure 1 As shown, the disclosed embodiment of the present invention provides a text-based banking business knowledge base processing device, comprising:

[0071] The acquisition unit is used to acquire the bank annual reports of at least two banks, and extract the asset business texts and liability business texts corresponding to the two banks respectively according to the two bank annual reports; wherein,

[0072] The asset business text includes a first asset business text and a second asset business text, and the liability business text includes a first liability business text and a second liability business text, the first asset business text and the first liability business text correspond to one of the two banks, and the second asset business text and the second liability business text correspond to the other of the two banks;

[0073] a conversion unit, configured to convert the first asset business text and the first liability business text into a first word vector, and to convert the second asset business text and the second liability business text into a second word vector, and the conversion unit is further configured to convert the first asset business text and the first liability business text into a first text data word list, and to convert the second asset business text and the second liability business text into a second text data word list;

[0074] A first calculation unit, used to calculate the correlation between the first word vector and the second word vector by cosine similarity;

[0075] A second calculation unit, used for calculating the association between the first text data word list and the second text data word list by Jaccard similarity;

[0076] A processing unit, used for determining the asset business association level between two banks according to the calculation results of the cosine similarity and the calculation results of the Jaccard similarity;

[0077] The prediction unit is used to predict the systemic risk level according to the asset business correlation level.

[0078] In a specific embodiment of the present application, a text acquisition module is provided in the prediction unit, and the text acquisition module is used to acquire litigation data according to big data technology, and the litigation data includes judgment document network data and litigation case data;

[0079] The prediction unit is also provided with a text extraction module, which is also used to extract keywords from the litigation data and predict the credit risk level according to the preset risk identification model. The risk identification model has several preset keywords and several preset credit risk levels. The preset credit risk levels are divided from high to low into: third-level credit risk level, second-level credit risk level, and first-level credit risk level; among them,

[0080] When the consistency rate between the keywords appearing in the litigation data and the preset keywords in the risk identification model is greater than or equal to 30%, the first-level credit risk level is output; when the consistency rate between the preset keywords in the risk identification model is greater than or equal to 60%, the second-level credit risk level is output; when the consistency rate between the preset keywords in the risk identification model is greater than or equal to 90%, the third-level credit risk level is output;

[0081] A text fine-tuning module is also provided in the prediction unit. The text fine-tuning module is used to extract the keywords appearing in the litigation data when the consistency rate between the keywords appearing in the litigation data and several keywords preset in the risk identification model is less than 30%, and add several preset keywords that do not exist in the risk identification model to the risk identification model according to the extraction results.

[0082] It should be noted that the present invention uses the data from the Judgment Documents Network and its own text data of case data for fine-tuning training based on the open source Baichuan big model, and extracts the subject matter, entity and relationship of the case based on keyword extraction technology, which is used to classify the risk level of the case, further enhance the credit risk control capabilities, and is suitable for multiple scenario applications of different text data.

[0083] In a specific embodiment of the present application, when several preset keywords that do not exist in the risk identification model are added to the risk identification model according to the extraction results, it also includes optimizing the risk identification model through Pre-Training, Fine-tuning, Prompt-tuning, and Instruction-tuning.

[0084] It should be noted that Pre-Training refers to the use of unsupervised learning methods on large-scale datasets to initialize the model in the field of machine learning. Through pre-training, the model can learn the common features and patterns of the data, thereby improving its performance on specific tasks. Pre-trained models are usually fine-tuned on specific fields or tasks to improve performance. Fine-Tuning refers to further training the model on the basis of the pre-trained model using labeled data for a specific task. The purpose of fine-tuning is to adjust the parameters of the model to adapt it to the requirements of a specific task, such as classification, language generation, etc. Fine-tuning usually requires less labeled data because the model has learned common features through pre-training. Prompt-Tuning refers to adjusting the behavior of the model by providing specific prompts or prompts to it. In the field of natural language processing, prompt learning is often used to guide the model to generate text that meets specific requirements or complete specific tasks. Through different prompts, the model can be guided to produce different outputs. Instruction-Tuning is a specific fine-tuning method that is often used for language generation models. In instruction learning, the model receives a form of instruction or control signal to guide it to generate a specific type of text, such as a sentence or paragraph that meets certain conditions. This method is often used to enhance the model's generation ability and control the quality of the generated results.

[0085] It is understandable that this method adopts the Fine-tuning fine-tuning method. Since the full-parameter fine-tuning method is time-consuming and computationally intensive and requires a large amount of hardware resources to support it, this method adopts the Prompt-tuning fine-tuning method, which only requires a small number of samples and a small number of GPU servers to support it.

[0086] It can be understood that by changing the traditional technology that relies on manual analysis of fields, the technical means of this application measures credit risk by storing it in a risk identification model for statistics. The risk identification model can be used to identify the overall risk situation, reducing the cost of manual analysis and improving business efficiency.

[0087] In a specific embodiment of the present application, the first calculation unit is used to calculate the correlation between the first word vector and the second word vector by the following formula:

[0088]

[0089] Where, s im-cos x,y is the correlation between the first word vector and the second word vector, which is used to determine the asset business correlation between the two banks. i is the first word vector, y i is the second word vector.

[0090] It can be understood that the cosine similarity value is used as a measure of the difference between the asset businesses of two banks. The specific meaning is: the cosine value is close to 1 and the angle tends to 0, indicating that the more similar the two word vectors are, the higher the correlation between the asset businesses of the two banks is; while the cosine value is close to 0 and the angle tends to 90 degrees, indicating that the more dissimilar the two word vectors are, the lower the correlation between the asset businesses of the two banks is.

[0091] In a specific embodiment of the present application, a first preset cosine similarity correlation degree T01, a second preset cosine similarity correlation degree T02, a third preset cosine similarity correlation degree T03, and a fourth preset cosine similarity correlation degree T04 are preset in the processing unit, and 0<T01<T02<T03<T04<1;

[0092] The processing unit is pre-set with a first preset asset-business correlation level A1, a second preset asset-business correlation level A2, a third preset asset-business correlation level A3, and a fourth preset asset-business correlation level A4, and A1<A2<A3<A4;

[0093] The processing unit is also used to select a corresponding asset business association level as the asset business association level between the two banks according to the relationship between the association calculation result j between the first word vector and the second word vector and each preset cosine similarity association level;

[0094] When j<T01, the first preset asset business association level A1 is selected as the asset business association level between the two banks;

[0095] When T01≤j<T02, the second preset asset business association level A2 is selected as the asset business association level between the two banks;

[0096] When T02≤j<T03, the third preset asset business association level A3 is selected as the asset business association level between the two banks;

[0097] When T03≤j<T04, the fourth preset asset-business association level A4 is selected as the asset-business association level between the two banks.

[0098] It can be understood that by pre-setting different cosine similarity correlations and asset business correlation levels, the accuracy of calculating the correlation between inter-bank asset business is improved. The processing unit can automatically select the appropriate asset business correlation level based on the correlation calculation result between the first word vector and the second word vector. This grading strategy makes the correlation calculation more refined, which helps to accurately judge the correlation between inter-bank business, thereby improving the efficiency and accuracy of financial services.

[0099] In a specific embodiment of the present application, the second calculation unit is used to calculate the association between the first text data word list and the second text data word list by the following formula:

[0100]

[0101] In the formula, s im-jaccard is the association between the first text data word list and the second text data word list, A is the set of the first text data word list, and B is the set of the second text data word list. The numerator represents the number of non-zero elements in the intersection of the two vectors, and the denominator represents the number of non-zero elements in the union of the two vectors.

[0102] It should be noted that Jaccard similarity calculation is a similarity analysis based on probability statistics, which specifically measures the difference between two sets. It does not consider the influence of word frequency, but only considers whether the word appears.

[0103] It can be understood that, assuming that the text data word lists of two bank asset businesses in the i-th year correspond to two sets A and B respectively, the Jaccard similarity value is used as a measure of the size of the difference between the two bank asset businesses. The specific meaning is: the more elements the two sets have in common, the more similar they are, and the higher the correlation between the two bank asset businesses.

[0104] In a specific embodiment of the present application, a first preset Jaccard similarity correlation K01, a second preset Jaccard similarity correlation K02, and a third preset Jaccard similarity correlation K03 are preset in the processing unit, and K01<K02<K03;

[0105] The processing unit is further configured to increase each preset asset business association level according to the relationship between the association calculation result i between the first text data word list and the second text data word list and each preset Jaccard similarity association level, and use the increased asset business association level as the asset business association level between the two banks;

[0106] When i<K01, and j<T01, the first preset asset business association level is upgraded by one level, and the second preset asset business association level is used as the asset business association level between the two banks;

[0107] When i<K01, and T01≤j<T02, the second preset asset business association level is upgraded by one level, and the third preset asset business association level is used as the asset business association level between the two banks;

[0108] When i<K01, and T02≤j<T03, the third preset asset business association level is upgraded by one level, and the fourth preset asset business association level is used as the asset business association level between the two banks;

[0109] When K01≤i<K02, and j<T01, the first preset asset business association level is raised by two levels, and the third preset asset business association level is used as the asset business association level between the two banks;

[0110] When K01≤i<K02, and T01≤j<T02, the second preset asset business association level is raised by two levels, and the fourth preset asset business association level is used as the asset business association level between the two banks;

[0111] When K02≤i<K03, and j<T01, the first preset asset business association level is upgraded by three levels, and the fourth preset asset business association level is used as the asset business association level between the two banks.

[0112] It can be understood that the above-mentioned technical means adjusts the business association level of each preset asset in real time according to the calculation results of the association between the text data word lists, so that the association level is more in line with the actual situation. By presetting different Jaccard similarity associations and adjusting the association level according to the actual calculation results, the accuracy of the system judgment is improved.

[0113] In a specific embodiment of the present application, the prediction unit is further used to calculate the systemic risk by the following formula:

[0114] mes i,t =α0+α1asset_cos i,t +α2lnscale i,t +α3fdirri i,t +α4epu i,t +α5npl i,t +α6ldr i,t +u i,t ;

[0115] In the formula, mes i,t is a systematic risk measurement indicator, asset-cos i,t is the asset business correlation between the two banks, lnscale i,t is the logarithm of total bank assets, epu i,t is the economic policy uncertainty index, fdirri i,t The foreign direct investment regulatory restriction index, npl i,t is the bank's non-performing loan ratio, idr i,tis the bank loan-to-deposit ratio, u i,t is the random disturbance term, a0 is the constant term, and a1 to a6 are regression coefficients.

[0116] It should be noted that a1 to a6 are regression coefficients used to reflect the impact of each variable on systemic risk.

[0117] In a specific embodiment of the present application, the prediction unit is further used to predict the systemic risk level according to the relationship between the calculated systemic risk value g and the preset systemic risk hazard value g0; wherein,

[0118] When g>g0, the first systemic risk level is taken as the predicted systemic risk level;

[0119] When g>1.2g0, the second systemic risk level is taken as the predicted systemic risk level;

[0120] When g>1.4g0, the third systemic risk level is taken as the predicted systemic risk level;

[0121] When g>1.6g0, the fourth systemic risk level is taken as the predicted systemic risk level.

[0122] In a specific embodiment of the present application, the prediction unit is further set with a first preset risk compensation fund L01, a second preset risk compensation fund L02, a third preset risk compensation fund L03, and a fourth preset risk compensation fund L04, and L01<L02<L03<L04;

[0123] The prediction unit is also used to determine the generated risk compensation fund according to the predicted systemic risk level;

[0124] When the first systemic risk level is used as the predicted systemic risk level, the first preset risk compensation fund L01 is determined as the generated risk compensation fund;

[0125] When the second systemic risk level is used as the predicted systemic risk level, the second preset risk compensation fund L02 is determined as the generated risk compensation fund;

[0126] When the third systemic risk level is used as the predicted systemic risk level, the third preset risk compensation fund L03 is determined as the generated risk compensation fund;

[0127] When the fourth systemic risk level is used as the predicted systemic risk level, the fourth preset risk compensation fund L04 is determined as the generated risk compensation fund.

[0128] It should be noted that the risk compensation fund is a financial instrument used to compensate investors for possible losses in certain investments. Usually, this fund provides certain compensation when investors suffer losses to reduce investors' risks.

[0129] It is understandable that the above technical solution helps to maintain the stability of the financial market, prevent financial crises caused by systemic risks, and protect the interests of financial institutions and investors. Setting different risk compensation funds based on the relationship between g and g0 is conducive to the refined management of risks and the generation of targeted risk response measures.

[0130] Based on the same technical concept, see Figure 2 As shown, the present invention also provides a text-based banking business knowledge base processing method, which is applied to a text-based banking business knowledge base processing device, comprising:

[0131] Obtain the bank annual reports of at least two banks, and extract the asset business texts and liability business texts corresponding to the two banks respectively according to the two bank annual reports; wherein,

[0132] The asset business text includes a first asset business text and a second asset business text, and the liability business text includes a first liability business text and a second liability business text, the first asset business text and the first liability business text correspond to one of the two banks, and the second asset business text and the second liability business text correspond to the other of the two banks;

[0133] The first asset business text and the first liability business text are converted into a first word vector, and the second asset business text and the second liability business text are converted into a second word vector, and the conversion unit is further used to convert the first asset business text and the first liability business text into a first text data word list, and convert the second asset business text and the second liability business text into a second text data word list;

[0134] Calculate the correlation between the first word vector and the second word vector through cosine similarity;

[0135] Calculating the association between the first text data word list and the second text data word list by Jaccard similarity;

[0136] The asset business correlation level between the two banks is determined based on the calculation results of cosine similarity and Jaccard similarity;

[0137] Predict the level of systemic risk based on the asset-business correlation level.

[0138] In summary, the present invention extracts and compares the asset business texts and liability business texts of different banks by processing and analyzing bank annual reports, thereby judging the business correlation between banks and further predicting the systemic risk level. The combination of automated processing and analysis greatly improves the efficiency of business correlation and risk prediction. By using a large text model, the business texts of banks can be extracted and compared more accurately, thereby obtaining more accurate correlation and risk levels. Moreover, by using machine learning and natural language processing technology, the judgment process of business correlation and risk level is made more intelligent. The present invention is applied to actual banking business operations, and can effectively help banks and management departments better understand and monitor the development trend of banking business, as well as discover and prevent potential risks in advance. The present invention has the advantages of high efficiency, accuracy and practicality.

[0139] The above is only an example of implementation of the present invention, but it cannot be used to limit the scope of the present invention. Any structural changes made according to the present invention should be regarded as falling within the protection scope of the present invention and subject to restrictions as long as they do not lose the essence of the present invention.

[0140] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0141] It should be noted that the system provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiments can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not regarded as improper limitations of the present invention.

[0142] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0143] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus / device.

[0144] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0145] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.

Claims

1. A text-based banking knowledge base processing device, characterized in that: include: The acquisition unit is used to acquire the bank annual reports of at least two banks, and extract the asset business texts and liability business texts corresponding to the two banks respectively according to the two bank annual reports; wherein, The asset business text includes a first asset business text and a second asset business text, and the liability business text includes a first liability business text and a second liability business text, wherein the first asset business text and the first liability business text correspond to one of the two banks, and the second asset business text and the second liability business text correspond to the other of the two banks; a conversion unit, configured to convert the first asset business text and the first liability business text into a first word vector, and to convert the second asset business text and the second liability business text into a second word vector, the conversion unit further configured to convert the first asset business text and the first liability business text into a first text data word list, and to convert the second asset business text and the second liability business text into a second text data word list; A first calculation unit, configured to calculate the correlation between the first word vector and the second word vector by using cosine similarity; A second calculation unit, configured to calculate the association between the first text data word list and the second text data word list by using Jaccard similarity; A processing unit, used for determining the asset business association level between the two banks according to the calculation results of the cosine similarity and the calculation results of the Jaccard similarity; A prediction unit, used for predicting the systemic risk level according to the asset-business correlation level; The prediction unit is provided with a text acquisition module, and the text acquisition module is used to acquire litigation data according to big data technology, and the litigation data includes judgment document network data and litigation case data; The prediction unit is also provided with a text extraction module, which is further used to extract keywords from the litigation data and predict the credit risk level according to a preset risk identification model. The risk identification model has a number of preset keywords and a number of preset credit risk levels. The preset credit risk levels are divided from high to low into: a third-level credit risk level, a second-level credit risk level, and a first-level credit risk level; wherein, When the consistency rate between the keywords appearing in the litigation data and the preset keywords in the risk identification model is greater than or equal to 30%, the first-level credit risk level is output; when the consistency rate between the preset keywords in the risk identification model is greater than or equal to 60%, the second-level credit risk level is output; when the consistency rate between the preset keywords in the risk identification model is greater than or equal to 90%, the third-level credit risk level is output; The prediction unit is also provided with a text fine-tuning module, which is used to extract the keywords appearing in the litigation data when the consistency rate between the keywords appearing in the litigation data and the preset keywords in the risk identification model is less than 30%, and add the preset keywords that do not exist in the risk identification model to the risk identification model according to the extraction results; The first calculation unit is used to calculate the association between the first word vector and the second word vector by using the following formula: ; Where sim-cos x,y is the correlation between the first word vector and the second word vector, which is used to determine the asset business correlation between the two banks. i is the first word vector, y i is the second word vector; The processing unit is pre-set with a first preset cosine similarity correlation degree T01, a second preset cosine similarity correlation degree T02, a third preset cosine similarity correlation degree T03, and a fourth preset cosine similarity correlation degree T04, and 0<T01<T02<T03<T04<1; The processing unit is pre-set with a first preset asset-business correlation level A1, a second preset asset-business correlation level A2, a third preset asset-business correlation level A3, and a fourth preset asset-business correlation level A4, and A1<A2<A3<A4; The processing unit is further configured to select a corresponding asset-business association level as the asset-business association level between the two banks according to the relationship between the association calculation result j between the first word vector and the second word vector and each preset cosine similarity association level; When j<T01, the first preset asset business association level A1 is selected as the asset business association level between the two banks; When T01≤j<T02, the second preset asset business association level A2 is selected as the asset business association level between the two banks; When T02≤j<T03, the third preset asset business association level A3 is selected as the asset business association level between the two banks; When T03≤j<T04, the fourth preset asset business association level A4 is selected as the asset business association level between the two banks; The second calculation unit is used to calculate the association between the first text data word list and the second text data word list by the following formula: ; Wherein, sim-jaccard is the association between the first text data word list and the second text data word list, A is the set of the first text data word list, B is the set of the second text data word list, wherein the numerator represents the number of non-zero elements in the intersection of the two vectors, and the denominator represents the number of non-zero elements in the union of the two vectors; The processing unit is pre-set with a first preset Jaccard similarity correlation K01, a second preset Jaccard similarity correlation K02, and a third preset Jaccard similarity correlation K03, and K01<K02<K03; The processing unit is further configured to increase each preset asset-business association level according to the relationship between the association calculation result i between the first text data word list and the second text data word list and each preset Jaccard similarity association level, and use the increased asset-business association level as the asset-business association level between the two banks; When i<K01, and j<T01, the first preset asset business association level is raised by one level, and the second preset asset business association level is used as the asset business association level between the two banks; When i<K01, and T01≤j<T02, the second preset asset business association level is raised by one level, and the third preset asset business association level is used as the asset business association level between the two banks; When i<K01, and T02≤j<T03, the third preset asset business association level is raised by one level, and the fourth preset asset business association level is used as the asset business association level between the two banks; When K01≤i<K02, and j<T01, the first preset asset business association level is raised by two levels, and the third preset asset business association level is used as the asset business association level between the two banks; When K01≤i<K02, and T01≤j<T02, the second preset asset business association level is raised by two levels, and the fourth preset asset business association level is used as the asset business association level between the two banks; When K02≤i<K03, and j<T01, the first preset asset business association level is upgraded by three levels, and the fourth preset asset business association level is used as the asset business association level between the two banks; The prediction unit is also used to calculate the systemic risk by the following formula: ; In the formula, mes i,t is a systematic risk measurement indicator, asset-cos i,t is the asset business correlation between the two banks, lnscale i,t is the logarithm of total bank assets, epu i,t is the economic policy uncertainty index, fdirri i,t The foreign direct investment regulatory restriction index, npl i,t is the bank's non-performing loan ratio, ldr i,t is the bank loan-to-deposit ratio, u i,t is the random disturbance term, a0 is the constant term, and a1 to a6 are regression coefficients; The prediction unit is also used to predict the systemic risk level according to the relationship between the calculated value g of the systemic risk and the preset systemic risk hazard value g0; wherein, When g>g0, the first systemic risk level is used as the predicted systemic risk level; When g>1.2g0, the second systemic risk level is used as the predicted systemic risk level; When g>1.4g0, the third systemic risk level is used as the predicted systemic risk level; When g>1.6g0, the fourth systemic risk level is taken as the predicted systemic risk level.

2. The text-based banking knowledge base processing device according to claim 1, characterized in that: The prediction unit is also set with a first preset risk compensation fund L01, a second preset risk compensation fund L02, a third preset risk compensation fund L03, and a fourth preset risk compensation fund L04, and L01<L02<L03<L04; The prediction unit is also used to determine the generated risk compensation fund according to the predicted systemic risk level; wherein, When the first systemic risk level is used as the predicted systemic risk level, determining the first preset risk compensation fund L01 as the generated risk compensation fund; When the second systemic risk level is used as the predicted systemic risk level, determining the second preset risk compensation fund L02 as the generated risk compensation fund; When the third systemic risk level is used as the predicted systemic risk level, determining the third preset risk compensation fund L03 as the generated risk compensation fund; When the fourth systemic risk level is used as the predicted systemic risk level, the fourth preset risk compensation fund L04 is determined as the generated risk compensation fund.

3. A text-based banking business knowledge base processing method, applied to a text-based banking business knowledge base processing device as claimed in any one of claims 1 to 2, characterized in that: include: Obtain bank annual reports of at least two banks, and extract asset business texts and liability business texts corresponding to the two banks respectively according to the two bank annual reports; wherein, The asset business text includes a first asset business text and a second asset business text, and the liability business text includes a first liability business text and a second liability business text, wherein the first asset business text and the first liability business text correspond to one of the two banks, and the second asset business text and the second liability business text correspond to the other of the two banks; According to the first asset business text and the first liability business text, the first word vector is converted, and according to the second asset business text and the second liability business text, the second word vector is converted, and the conversion unit is further used to convert the first asset business text and the first liability business text into a first text data word list, and according to the second asset business text and the second liability business text into a second text data word list; Calculate the correlation between the first word vector and the second word vector by cosine similarity; Calculating the association between the first text data word list and the second text data word list by Jaccard similarity; Determining the asset business correlation level between the two banks according to the calculation results of the cosine similarity and the calculation results of the Jaccard similarity; The systemic risk level is predicted based on the asset-business correlation level.