CN119091454A

CN119091454A - A method, system, device and medium for identifying abnormal information of financial electronic bills

Info

Publication number: CN119091454A
Application number: CN202411591474.8A
Authority: CN
Inventors: 陈庸凯; 周丽红; 黄荣明; 陈海量; 池文倩; 马作玲
Original assignee: Fujian Boss Software Co ltd
Current assignee: Fujian Boss Software Co ltd
Priority date: 2024-11-08
Filing date: 2024-11-08
Publication date: 2024-12-06

Abstract

The invention relates to a financial electronic bill abnormal information identification method, a system, equipment and a medium, wherein the method comprises the following steps of collecting bill data, filtering invalid or wrong bill data, and extracting key characteristics of the filtered bill data; the method comprises the steps of obtaining reinforced bill data from bill data by using a data reinforcing technology, converting the reinforced bill data into a data set in a JSON format, carrying out fine tuning training on a Qwen-7b large language model by using the data set, designing fine tuning training prompt words aiming at key information in the bill to guide the Qwen-7b large language model to pay attention to and extract the information better, judging the financial electronic bill based on the trained Qwen-7b large language model, and outputting an identification result to realize monitoring of financial electronic bill item channeling and identification of abnormal information.

Description

Financial electronic bill abnormal information identification method, system, equipment and medium

Technical Field

The invention relates to a method, a system, equipment and a medium for identifying abnormal information of a financial electronic bill, belonging to the field of financial electronic bill supervision.

Background

With the development of information technology, financial electronic bills have become an important component of modern financial management. They not only improve the efficiency of bill handling, but also reduce the use of paper bills and reduce the possibility of human error. However, with the popularization of electronic bills, detection and processing of abnormal information are also a key problem. Abnormal information including forgery, tampering, resubmitting, expiration of notes, etc. may lead to loss of financial funds and confusion of financial management if not found in time.

At present, the conventional methods for identifying abnormal information mainly comprise a rule-based method and a statistical method. Rule-based methods screen notes by preset rules, which are typically based on experience and historical data. However, this approach has difficulty coping with complex and variable anomaly patterns and new fraudulent means. Statistical methods find abnormal patterns through data analysis, but these methods generally require a large amount of historical data and high computing power, and have limited adaptability to real-time and dynamic changes.

The prior art, such as U.S. patent No. US10795752B2, discloses a data validation method. The method includes the steps of extracting journal entry data and supporting documents to be verified, including reconciliation to identify potential errors. And extracting the entity related to the ledger by using a natural language processing technology, and determining the value of the corresponding data field and the mapping relation thereof. And if the one-to-one mapping exists, selecting a corresponding value, otherwise, acquiring enhanced data and processing to determine the value. The value is compared to the actual value, generating a potential error notification. By processing the data through rules and machine learning, anomalies are detected and contextual information of behavioral deviations is identified.

The problem with the prior art described above is that it is highly dependent on the accuracy of the NLP technique and machine learning models, the performance of which may be affected by a number of factors, such as data quality, model design, training data, etc. This approach may be very effective in processing structured or semi-structured data by processing the data through detailed classification and mapping logic. However, this approach may become inflexible when faced with highly unstructured or frequently changing data.

Disclosure of Invention

In order to solve the problems in the prior art, the invention provides a method, a system, equipment and a medium for identifying abnormal information of a financial electronic bill.

The technical scheme of the invention is as follows:

in one aspect, the invention provides a method for identifying abnormal information of a financial electronic bill, which comprises the following steps:

Collecting bill data, filtering invalid or wrong bill data in a format, and extracting key characteristics of the filtered bill data, wherein the bill data comprises financial policy regulations, laws and regulations, policy interpretation and historical bill data, and the key characteristics comprise billing units, bill types, bill total amount, bill items, item amount, remark information and secondary item information;

The reinforced bill data is obtained by using a data reinforcing technology, and the reinforced bill data is converted into a data set in a JSON format;

performing fine tuning training on the Qwen-7b large language model by using a data set, and designing fine tuning training prompt words aiming at key information in the bill so as to guide the Qwen-7b large language model to pay attention to and extract the information better;

And judging the financial electronic bill based on the trained Qwen-7b large language model, and outputting the identification result.

As a preferred embodiment, the data enhancement technique includes transliteration, paraphrasing, substitution, and cartesian collection;

The translation is carried out, namely remarks or secondary project information in bill data is translated into other languages through a translation tool for multiple times, and the other languages are translated back into Chinese, so that similar expression of sentences is obtained;

The substitution of the paraphrasing words, namely, substituting remarks in bill data or keywords in secondary project information by the paraphrasing words to form a new sample, wherein the keywords comprise, but are not limited to, donation, sponsorship and funding;

and D, the Cartesian set is used for making the tamper suspicious data in the bill data and the random bill data into a new suspicious sample.

As a preferred embodiment, the fine tuning training method of the Qwen-7b large language model comprises the following steps:

dividing the data set in the JSON format into a training data set and a verification data set;

using Transformers libraries to load Qwen-7b models;

configuring the learning rate, batch size and training wheel of training;

Loading a pre-trained QLora model;

writing training circulation, performing fine adjustment on the Qwen-7b model by using the QLora model and the training data set, and recording the weight in the fine adjustment process;

and evaluating the performance of the trimmed model by using the verification set, and adjusting the weight in the trimming process based on the performance of the model.

As a preferred implementation mode, RAG technology is combined in the model reasoning process, specifically, information in a relevant knowledge base or a database is searched through the RAG technology, and the generation capacity of the model is enhanced.

As a preferred embodiment, the prompt word includes a project name and remark content;

The prompt word is stored in a database, and is injected into Qwen-7b model through a preprocessing statement of MyBatis framework.

On the other hand, the invention also provides a financial electronic bill abnormal information identification system, which comprises:

The data acquisition module acquires bill data, filters invalid or wrong bill data in a format and extracts key characteristics of the filtered bill data, wherein the bill data comprises financial policy regulation, law and regulation, policy interpretation and historical bill data, and the key characteristics comprise billing units, bill types, bill total amount, bill items, item amounts, remark information and secondary item information;

The data processing module is used for obtaining reinforced bill data by using a data reinforcing technology, and converting the reinforced bill data into a data set in a JSON format;

The model training module is used for carrying out fine tuning training on the Qwen-7b large language model by using the data set, and designing fine tuning training prompt words aiming at key information in the bill so as to guide the Qwen-7b large language model to pay attention to and extract the information better;

and the result output module is used for judging the financial electronic bill based on the trained Qwen-7b large language model and outputting the identification result.

In yet another aspect, the present invention further provides an electronic device, on which a computer program is stored, which when executed by a processor implements a method for identifying financial electronic ticket anomaly information according to any embodiment of the present invention.

In yet another aspect, the present invention also provides a computer readable medium storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the method for identifying financial electronic ticket anomaly information according to any embodiment of the present invention.

The invention has the following beneficial effects:

The invention adopts the data enhancement technology to process bill data, which is helpful to increase the diversity and the richness of the data set, thereby improving the accuracy and the robustness of the subsequent model training. By data enhancement, the model can learn more diversified data features, thereby performing more excellently when processing actual data. Fine tuning training is performed using a large language model Qwen-7b that has powerful language understanding and generation capabilities. By designing the fine tuning training prompt words aiming at key information in the bill, the model can be guided to pay attention to and extract the information better, and the accuracy and efficiency of data processing are further improved.

The automatic processing and judgment of bill data are realized, and the burden of manual auditing is greatly reduced. The finance electronic bill can be rapidly and accurately identified and judged through the trained Qwen-7b large language model, and an identification result is output. This not only improves the working efficiency, but also reduces the risk of human error.

Drawings

FIG. 1 is a flow chart of a method according to an embodiment of the invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

It should be understood that the step numbers used herein are for convenience of description only and are not limiting as to the order in which the steps are performed.

It is to be understood that the terminology used in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

The terms "comprises" and "comprising" indicate the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

The term "and/or" refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.

Embodiment one:

referring to fig. 1, the invention provides a method for identifying abnormal information of a financial electronic bill, which comprises the following steps:

Data augmentation (Data Augmentation), also known as data augmentation, is a technique whereby training sets are artificially augmented by using existing data to create modified copies of the data set, which is one of the techniques commonly used in deep learning, including making minor changes to the data set or using deep learning to generate new data points. The data enhancement is mainly used for increasing training data sets, so that the data sets are diversified as much as possible, and the trained model has stronger generalization capability.

Because the suspicious data set is smaller, the suspicious bill is mainly subjected to data enhancement, so that the data of the suspicious bill and the normal bill are basically balanced.

using Transformers libraries to load Qwen-7b models;

configuring the learning rate, batch size and training wheel of training;

Loading a pre-trained QLora model;

Qwen-7b characteristics the model is based on a transducer structure, has 70 hundred million parameters and is excellent in a plurality of natural language processing tasks. Its powerful text generation and understanding capabilities provide a solid basis for the identification of fiscal suspicious notes.

This approach helps to prevent SQL injection attacks because it sets the parameter values by preprocessing the statement (PREPARED STATEMENTS) instead of directly splicing the parameter values into the SQL statement.

Embodiment two:

the invention also provides a financial electronic bill abnormal information identification system, which comprises:

Embodiment III:

the present embodiment provides an electronic device, on which a computer program is stored, which when executed by a processor implements a financial electronic ticket anomaly information identification method according to any one of the embodiments of the present invention.

Embodiment four:

The present embodiment provides a computer-readable medium storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the financial electronic ticket anomaly information identification method according to any one of the embodiments of the present invention.

In the embodiments of the present application, "at least one" means one or more, and "a plurality" means two or more. "and/or", describes an association relation of association objects, and indicates that there may be three kinds of relations, for example, a and/or B, and may indicate that a alone exists, a and B together, and B alone exists. Wherein A, B may be singular or plural. The character "/" generally indicates that the context-dependent object is an "or" relationship. "at least one of the following" and the like means any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c may represent a, b, c, a and b, a and c, b and c, or a and b and c, wherein a, b, c may be single or plural.

Those of ordinary skill in the art will appreciate that the various elements and algorithm steps described in the embodiments disclosed herein can be implemented as a combination of electronic hardware, computer software, and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, and are not repeated herein.

In several embodiments provided by the present application, any of the functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application. The storage medium includes various media capable of storing program codes, such as a U disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk.

The foregoing description is only illustrative of the present invention and is not intended to limit the scope of the invention, and all equivalent structures or equivalent processes or direct or indirect application in other related technical fields are included in the scope of the present invention.

Claims

1. A financial electronic bill abnormal information identification method is characterized by comprising the following steps of collecting bill data, filtering invalid or wrong bill data in a format, extracting key features of the filtered bill data, wherein the bill data comprise financial policy regulations, law regulations, policy interpretation and historical bill data, the key features comprise billing units, bill types, bill total amount, bill items, item amounts, remark information and secondary item information, the bill data are enhanced by using a data enhancement technology, the enhanced bill data are converted into a data set in a JSON format, fine tuning training is conducted on a Qwen-7b large language model by using the data set, fine tuning training prompt words aiming at key information in the bill are designed to guide the Qwen-7b large language model to pay attention to and extract the information better, and the financial electronic bill is judged on the basis of the Qwen-7b large language model after training and an identification result is output.

2. The method for identifying financial electronic bill anomaly information according to claim 1 wherein the data enhancement technique includes back-translation, hyponym substitution and cartesian collection;

3. The method for identifying abnormal information of financial electronic bill according to claim 1, wherein the fine tuning training method of Qwen-7b big language model is as follows:

using Transformers libraries to load Qwen-7b models;

configuring the learning rate, batch size and training wheel of training;

Loading a pre-trained QLora model;

4. The method for identifying abnormal information of financial electronic bill according to claim 3, wherein the generation capacity of the model is enhanced by combining RAG technology in the model reasoning process, in particular by retrieving information in a related knowledge base or database through the RAG technology.

5. The method for identifying abnormal information of financial electronic bill according to claim 1, wherein the prompt word includes item name and remark content;

6. A financial electronic ticket anomaly information recognition system, comprising:

7. The system for identifying abnormal information of financial electronic bill according to claim 6 wherein the model training module, the fine tuning training method of Qwen-7b big language model is:

using Transformers libraries to load Qwen-7b models;

configuring the learning rate, batch size and training wheel of training;

Loading a pre-trained QLora model;

8. The system for identifying abnormal information of financial electronic bill according to claim 7 wherein the model training module combines with the RAG technology in the model reasoning process, in particular, the information in the relevant knowledge base or database is searched by the RAG technology, so as to enhance the generating capability of the model.

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the financial electronic ticket anomaly information identification method of any one of claims 1 to 5 when the program is executed by the processor.

10. A computer-readable storage medium having stored thereon a computer program, wherein the program when executed by a processor implements the financial electronic ticket anomaly information identification method of any one of claims 1 to 5.