Data analysis system and method based on natural language processing technology

Abstract

The invention relates to the field of intelligent analysis, and provides a data analysis system and method based on a natural language processing technology, and the method comprises the steps: firstly obtaining a to-be-audited business contract through a contract management system, and then carrying out the content semantic analysis of the to-be-audited business contract through a data analysis technology based on a natural language, and whether a risk exists in the business contract to be audited or not is automatically judged based on the analyzed features. Therefore, the problems of low efficiency, strong subjectivity and easy risk point omission in traditional contract risk management can be effectively solved.

Description

Technical Field

[0001] The present application relates to the field of intelligent analysis, and more specifically, to a data analysis system and method based on natural language processing technology. Background Art

[0002] Contracts are an important carrier of commercial activities and economic exchanges, which clearly define the rights and obligations of both parties, especially regarding the transaction amount, payment method, delivery conditions, etc. Through contract risk assessment, economic losses caused by breach of contract by the other party or defects in contract terms can be avoided. Traditional contract risk management often relies on manual review, which requires a lot of manpower and time to check the contract content word by word, and is relatively inefficient. In addition, when manually assessing contract risks, different auditors may have differences in professional knowledge, experience, personal understanding, etc., which may lead to strong subjectivity in risk assessment, and it is difficult to achieve complete unification of the assessment criteria. In addition, traditional methods only look for contract risks based on some common risk points or the auditors' past experience, which can easily miss some hidden or newly emerging risk points.

[0003] Therefore, a data analysis solution based on natural language processing technology is needed. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present application provides a data analysis system and method based on natural language processing technology.

[0005] A data analysis method based on natural language processing technology, comprising:

[0006] Obtain the business contracts to be reviewed from the contract management system;

[0007] Performing preliminary data processing on the business contract to be reviewed to obtain a plurality of business contract content word embedding vectors;

[0008] Performing semantic analysis on the multiple business contract content word embedding vectors to obtain a business contract content semantic understanding feature vector;

[0009] Based on the semantic understanding feature vector of the business contract content, an audit classification result of the business contract to be audited is obtained.

[0010] A data analysis system based on natural language processing technology, comprising:

[0011] The data collection module for business contracts to be reviewed is used to obtain business contracts to be reviewed from the contract management system;

[0012] A preliminary processing module for business contract data to be reviewed, used for preliminary processing of the data of the business contract to be reviewed to obtain a plurality of business contract content word embedding vectors;

[0013] A semantic analysis module for the business contract data to be reviewed, used to perform semantic analysis on the multiple business contract content word embedding vectors to obtain a business contract content semantic understanding feature vector;

[0014] The audit result generation module is used to obtain the audit classification result of the business contract to be audited based on the semantic understanding feature vector of the business contract content.

[0015] This application has significant technical effects due to the adoption of the above technical solutions:

[0016] The data analysis system and method based on natural language processing technology provided by the present application first obtains the business contract to be reviewed through the contract management system, then uses the data analysis technology based on natural language to perform content semantic analysis on the business contract to be reviewed, and automatically determines whether the business contract to be reviewed has risks based on the features obtained by the analysis. In this way, the problems of low efficiency, strong subjectivity and easy omission of risk points in traditional contract risk management can be effectively solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 The present invention is a flowchart of a data analysis method based on natural language processing technology according to an embodiment of the present application.

[0019] Figure 2 This is a flowchart of step S120 of the data analysis method based on natural language processing technology according to an embodiment of the present application.

[0020] Figure 3 This is a flowchart of step S130 of the data analysis method based on natural language processing technology according to an embodiment of the present application.

[0021] Figure 4 It is a system block diagram of a data analysis system based on natural language processing technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0023] As mentioned in the above background technology, as the cornerstone of commercial activities and economic exchanges, the contract is not only a written expression of the cooperation intention of both parties, but also a key legal document to ensure the smooth progress of the transaction and safeguard the rights and interests of both parties. It specifies in detail the specific amount of the transaction, payment methods (such as advance deposit, installment payment, letter of credit, etc.), delivery conditions (including time, place, quality standards and acceptance process) and other core elements, providing clear guidance for the cooperation framework of both parties.

[0024] In actual operation, the importance of contract risk assessment is self-evident. A good contract risk management system can effectively prevent economic losses and legal disputes caused by the other party's breach of contract (such as unjustified delay in payment, provision of products or services that do not meet the agreement) or defects in the contract terms themselves (such as vague expressions, omission of key terms). However, traditional contract risk management methods mostly rely on manual review, which not only means that a large amount of human resources are required, but also professionals review the contract text word by word, which is time-consuming and laborious, and the efficiency is often not high, especially when faced with a large number of contracts or complex contracts.

[0025] More importantly, when manually reviewing contract risks, the risk judgment is often highly subjective due to the different professional knowledge backgrounds, work experience, personal understanding ability, and familiarity with specific industries or markets of the reviewers. Different reviewers may have very different risk assessment results for the same contract based on their own understanding and experience, making it difficult to unify the evaluation criteria and increasing the uncertainty of corporate decision-making.

[0026] In addition, traditional methods of finding contract risks are often limited to some common risk points or the past experience of auditors. Although this "empiricism" approach can identify some common risks to a certain extent, it is easy to ignore those hidden deeper or newly emerging risk points. Therefore, a data analysis solution based on natural language processing technology is expected.

[0027] Figure 1 Flowchart of a data analysis method based on natural language processing technology according to an embodiment of the present application. Figure 1 As shown, according to the data analysis method based on natural language processing technology of the embodiment of the present application, it includes: S110, obtaining the business contract to be reviewed from the contract management system; S120, performing preliminary data processing on the business contract to be reviewed to obtain multiple business contract content word embedding vectors; S130, performing semantic analysis on the multiple business contract content word embedding vectors to obtain a business contract content semantic understanding feature vector; S140, based on the business contract content semantic understanding feature vector, obtaining the review classification result of the business contract to be reviewed.

[0028] In step S110, the business contract to be reviewed is obtained from the contract management system. It should be understood that the business contract to be reviewed contains basic contract information such as contract number, contract signing date, information of both parties to the contract, core transaction terms such as transaction amount, transaction content description, payment method, delivery conditions, breach of contract liability clauses and dispute resolution clauses. Specifically, the basic contract information builds a basic framework for overall risk judgment. For example, the contract signing date indicates the risk in the time dimension, and the information of both parties to the contract helps to assess the risks related to the cooperation object. These are the prerequisite factors that need to be considered in the subsequent in-depth analysis of the risks of specific terms, and assist in judging the risk situation of this contract in the entire business environment from a macro perspective. The core terms of the transaction directly involve the key links such as the substantive content and capital flow in the process of contract performance. The transaction amount determines the magnitude of the risk. Unclear description of the transaction content will lead to the risk of performance disputes. Unreasonable payment methods and delivery conditions are more likely to lead to direct breach of contract risks. Accurate analysis of these terms can lock in the main risk hazards in each key link of contract performance, which is the core focus of judging whether there is risk in the contract. A clear and reasonable breach of contract liability clause can, to a certain extent, constrain both parties to the contract to act in accordance with the contract and reduce the possibility of breach of contract. Once a breach of contract occurs, it is an important basis for the injured party to recover losses. Whether its setting is perfect is closely related to the overall risk of the contract. If this part is weak, it will make it difficult to control the losses after the risk occurs, which will affect the judgment of the degree of contract risk. Although the dispute resolution clause mainly acts on the handling stage after the risk event occurs, its rationality determines the cost and efficiency of the subsequent dispute resolution. If the dispute resolution clause is not conducive to efficient problem resolution, the negative impact of the risk will continue to expand in the future, indirectly reflecting the overall risk status of the contract, and it also needs to be taken into consideration when judging the contract risk. In general, the various content sections in the business contract to be reviewed are interrelated and influence each other, and together constitute a comprehensive basis for judging whether there is risk in the audited business contract.

[0029] In step S120, the business contract to be reviewed is preliminarily processed to obtain a plurality of business contract content word embedding vectors. Specifically, Figure 2 FIG. 1 is a flowchart of step S120 of the data analysis method based on natural language processing technology according to an embodiment of the present application. Figure 2 As shown, it includes: S121, preprocessing the business contract to be reviewed to obtain a preprocessed business contract; S122, segmenting the preprocessed business contract and passing it through a word embedding layer to obtain the word embedding vectors of the multiple business contract contents.

[0030] In step S121, the business contract to be reviewed is preprocessed to obtain a preprocessed business contract. It should be understood that the contract document contains information in different formats, and these different information formats will interfere with the subsequent natural language processing steps. At the same time, the acquired contract also contains some information that is not critical to risk judgment, such as headers, footers, some auxiliary instructions for the numbering system, notes on the stamping and signature areas, etc. If these irrelevant information are not processed, the amount of data processed will increase and the subsequent semantic analysis model may be confused, so that the model needs to spend extra energy to process these irrelevant contents, thereby affecting the efficiency and accuracy of risk judgment. Therefore, in the technical solution of the present application, it is necessary to preprocess the business contract to be reviewed. Through preprocessing, the business contract to be reviewed can be converted into a unified format that is convenient for subsequent processing, such as a plain text format, so that the processed text is consistent in subsequent steps such as word segmentation and word embedding, avoiding errors caused by format differences. At the same time, through preprocessing, information that is not directly related to contract risk judgment is deleted, so that the data can be more focused on the key terms of the contract, such as transaction content, payment method, breach of contract liability, etc. In this way, it helps to improve the specificity of subsequent semantic analysis and enables the model to more efficiently extract and analyze semantic features related to risks.

[0031] In step S122, the pre-processed business contract is segmented and then passed through the word embedding layer to obtain the multiple business contract content word embedding vectors. It should be understood that computers cannot directly understand the meaning of natural language texts like humans. For complex natural language content such as contract text, it is necessary to decompose it into basic units that can be processed by computers. Word segmentation is the key step in this decomposition process. It can convert continuous text streams into discrete vocabulary units, thereby providing a basis for subsequent more complex semantic analysis and processing. That is, by performing word segmentation operations on the pre-processed business contract, subsequent data processing can be facilitated. Next, the pre-processed business contract after word segmentation is input into the word embedding layer to obtain multiple business contract content word embedding vectors. It needs to be explained that the word embedding layer is a structure that maps vocabulary to a low-dimensional vector space, in which the semantic and grammatical information of vocabulary is represented in the form of vectors. For example, the positions of "buy" and "procurement", two words with similar semantics, in the word embedding vector space will be relatively close, and their vector representations will have certain similarities. The word embedding vectors generated by mapping the vocabulary to the vector space contain the semantic information of the vocabulary. These vectors can be used as input for subsequent semantic capturers to help mine deeper semantic understanding features of the contract text.

[0032] In step S130, semantic analysis is performed on the multiple business contract content word embedding vectors to obtain a business contract content semantic understanding feature vector. Specifically, Figure 3 FIG. 1 is a flow chart of step S130 of the data analysis method based on natural language processing technology according to an embodiment of the present application. Figure 3 As shown, the step S130 includes: S131, performing short-distance semantic analysis on the multiple business contract content word embedding vectors to obtain a short-distance semantic understanding feature vector of the business contract content; S132, performing long-distance semantic analysis on the multiple business contract content word embedding vectors to obtain a long-distance semantic understanding feature vector of the business contract content; S133, performing feature scale adaptation based on target domain association matching on the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content to obtain the semantic understanding feature vector of the business contract content.

[0033] In step S131, a short-distance semantic analysis is performed on the multiple business contract content word embedding vectors to obtain a short-distance semantic understanding feature vector of the business contract content. Specifically, in an embodiment of the present application, the step S131 includes: passing the multiple business contract content word embedding vectors through a first distance semantic capturer to obtain a short-distance semantic understanding feature vector of the business contract content. It should be understood that the contract text contains various clauses and expressions, among which short-distance semantic relationships are crucial to understanding the details of the contract. For example, in a sentence or several adjacent sentences, such as "the payment method is installment payment, the amount of each installment is one-third of the total contract amount, and the first installment payment time is within 10 working days after the contract is signed", the short-distance semantic association between these words directly reflects the specific details of the payment method. It is difficult to fully capture this complex local semantic relationship by relying solely on word embedding vectors. Based on this, in the technical solution of the present application, it is necessary to input the multiple business contract content word embedding vectors into the first distance semantic capturer. The first distance semantic capturer in the present application specifically refers to a recurrent neural network model. When processing input data, the recurrent neural network model can consider the order information of the word embedding vector in the sequence, and then understand the semantic combination within a short distance, so as to extract the semantic features between adjacent words or relatively close words in the original business contract text data to be reviewed, which can provide a basis for local semantic understanding for subsequent risk judgment.

[0034] In step S132, the multiple business contract content word embedding vectors are subjected to long-distance semantic analysis to obtain a long-distance semantic understanding feature vector of the business contract content. Specifically, in an embodiment of the present application, the step S132 includes: passing the multiple business contract content word embedding vectors through a second distance semantic capturer to obtain a long-distance semantic understanding feature vector of the business contract content. The second distance semantic capturer in the present application specifically refers to a bidirectional long short-term memory neural network model. It should be understood that a contract is a holistic text with internal logical coherence. In the contract text, a lot of important semantic information is reflected through long-distance relationships across sentences, paragraphs, and even chapters. For example, the beginning of the contract may mention the basic information of the two parties to the transaction and the overall goal of the transaction, while the middle part elaborates on the transaction method, payment terms, etc., and the last part is the liability for breach of contract and dispute resolution. There are long-distance semantic associations between these different parts, which are crucial for a comprehensive understanding of the contract and risk assessment. Long-distance semantic associations can reveal the overall intention and potential risks of the contract. Traditional methods are difficult to effectively capture long-distance semantics. As the text distance increases, the transmission and association of semantic information becomes complex, and simple models cannot handle such long-distance relationships well. The bidirectional long short-term memory neural network model has the ability to process long sequence data and can process text sequences in both the forward and reverse directions, thereby better capturing long-distance semantic dependencies. This feature makes it very suitable for mining long-distance semantic associations in contract texts. Therefore, in the technical solution of the present application, it is necessary to input the multiple business contract content word embedding vectors into the bidirectional long short-term memory neural network model to obtain the long-distance semantic understanding feature vector of the business contract content.

[0035] In step S133, the feature scale adaptation of the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content is performed based on the target domain association matching to obtain the semantic understanding feature vector of the business contract content. It should be understood that the semantic information contained in the contract text is complex and multi-dimensional, including both the semantics reflected by the local, short-distance clause details and the long-distance semantics that span a large length and are interrelated between clauses. A single short-distance semantic understanding feature vector can only reflect the relationship between the elements within a specific clause in the contract, while a single long-distance semantic understanding feature vector focuses on the long-distance association between different parts of the clauses. In order to more comprehensively and accurately understand the overall semantics of the contract and judge the risks, the technical solution of this application needs to integrate the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content. Through fusion, a complete semantic understanding foundation covering the contract text from local details to the overall structure can be constructed, which helps to more accurately judge whether there are risks in the contract.

[0036] In particular, considering that the short-distance semantic understanding feature vector of the business contract content is obtained by processing a recurrent neural network model, and the long-distance semantic understanding feature vector of the business contract content is obtained by processing a bidirectional long short-term memory neural network model, the difference in the structure of the two models will lead to different focuses on semantic capture, which will in turn lead to the generated features in the feature space of different construction logics. Their dimensional meanings, numerical distributions, etc. are quite different. If the fusion is performed directly, it is like forcibly merging vectors from different coordinate systems. It is difficult to accurately consider the comprehensive semantic contribution of each dimension after fusion and the complex nonlinear relationship between them, which makes it easy to ignore those more complex semantic relationships hidden at the intersection of different feature space dimensions that require deep mining to discover. If the complex relationship between the two feature vectors is ignored, the contract risk will be misjudged due to insufficient information utilization. Therefore, in the technical solution of the present application, it is necessary to adapt the feature scale of the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content based on the target domain association matching to obtain the semantic understanding feature vector of the business contract content.

[0037] Specifically, in an embodiment of the present application, the feature scale adaptation of the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content is performed based on target domain association matching to obtain the business contract content semantic understanding feature vector, including: mapping the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content to an intrinsic decomposition space to obtain the short-distance semantic understanding intrinsic feature vector of the business contract content and the long-distance semantic understanding intrinsic feature vector of the business contract content; performing feature scale adaptation of the short-distance semantic understanding intrinsic feature vector of the business contract content and the long-distance semantic understanding intrinsic feature vector of the business contract content The target domain-oriented feature scale is adjusted to obtain the modulated intrinsic feature vector of the short-distance semantic understanding of the business contract content and the modulated intrinsic feature vector of the long-distance semantic understanding of the business contract content; a fine-grained interactive semantic matrix of the semantic understanding of the business contract content between the modulated intrinsic feature vector of the short-distance semantic understanding of the business contract content and the modulated intrinsic feature vector of the long-distance semantic understanding of the business contract content is constructed; based on the fine-grained interactive semantic matrix of the semantic understanding of the business contract content, the modulated intrinsic feature vector of the short-distance semantic understanding of the business contract content and the modulated intrinsic feature vector of the long-distance semantic understanding of the business contract content are fused to obtain the semantic understanding feature vector of the business contract content.

[0038] More specifically, in an embodiment of the present application, a fine-grained interactive semantic matrix of the semantic understanding of the business contract content is constructed between the modulated intrinsic feature vector of the short-distance semantic understanding of the business contract content and the modulated intrinsic feature vector of the long-distance semantic understanding of the business contract content, including: multiplying the transposed vector of the modulated intrinsic feature vector of the short-distance semantic understanding of the business contract content and the modulated intrinsic feature vector of the long-distance semantic understanding of the business contract content and dividing the result by the scale of the modulated intrinsic feature vector of the long-distance semantic understanding of the business contract content to obtain a fine-grained interactive matrix of the semantic understanding of the business contract content; performing a convolution coding operation based on a convolution kernel of 3×3 on the fine-grained interactive matrix of the semantic understanding of the business contract content to obtain a fine-grained interactive semantic matrix of the semantic understanding of the business contract content.

[0039] More specifically, in an embodiment of the present application, the modulated intrinsic feature vector of the short-distance semantic understanding of the business contract content and the modulated intrinsic feature vector of the long-distance semantic understanding of the business contract content are fused based on the fine-grained interactive semantic matrix of the semantic understanding of the business contract content to obtain the semantic understanding feature vector of the business contract content, including: multiplying the fine-grained interactive semantic matrix of the semantic understanding of the business contract content by the modulated intrinsic feature vector of the short-distance semantic understanding of the business contract content to obtain the short-distance semantic understanding feature vector of the business contract content after information field mapping; multiplying the fine-grained interactive semantic matrix of the semantic understanding of the business contract content by the modulated intrinsic feature vector of the long-distance semantic understanding of the business contract content to obtain the long-distance semantic understanding feature vector of the business contract content after information field mapping; weighted fusion of the short-distance semantic understanding feature vector of the business contract content after information field mapping and the long-distance semantic understanding feature vector of the business contract content after information field mapping to obtain the semantic understanding feature vector of the business contract content.

[0040] In the embodiment of the present application, specifically, the feature scale adaptation based on the target domain association matching is performed on the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content to obtain the business contract content semantic understanding feature vector, including: processing the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content with the following formula to obtain the business contract content semantic understanding feature vector; wherein, the formula is:

[0041]

[0042] v c =αv kt +βv st

[0043] Where V1 represents the feature vector of short-distance semantic understanding of business contract content, V2 represents the feature vector of long-distance semantic understanding of business contract content, PCA(·) represents principal component analysis, U1 represents the sequence of eigendecomposition vectors of short-distance semantic understanding of business contract content, Λ1 represents the diagonal matrix of short-distance semantic understanding of business contract content, λ 11 and λ 1m denote the eigenvalues ​​of the first and mth positions of the short-distance semantic understanding diagonal matrix of the business contract content, T denotes the transpose of the vector, concat{·;·} denotes vector concatenation, and v 11 、v 12 、v 1m They represent the first, second and mth eigenvectors of the sequence of short-distance semantic understanding eigendecomposition vectors of the business contract content, V k represents the intrinsic feature vector of short-distance semantic understanding of business contract content, U2 represents the sequence of intrinsic decomposition vectors of long-distance semantic understanding of business contract content, Λ2 represents the diagonal matrix of short-distance semantic understanding of business contract content, λ 21 and λ 2m They represent the eigenvalues ​​of the first and mth positions of the diagonal matrix of the long-distance semantic understanding of the business contract content, respectively. 21 、v 22 、v 2m They represent the first, second and mth eigenvectors of the sequence of eigendecomposition vectors of the long-distance semantic understanding of the business contract content, V s Represents the intrinsic feature vector of long-distance semantic understanding of business contract content, v i The i-th eigenvalue of the eigenvector representing the short-distance semantic understanding of the business contract content, v j The jth eigenvalue of the eigenvector representing the long-distance semantic understanding of the business contract content, represents the square of the vector norm, exp represents the natural exponential function, V′ k Represents the intrinsic feature vector after short-distance semantic understanding of the business contract content, V′ s Represents the intrinsic feature vector after long-distance semantic understanding modulation of the business contract content, represents matrix multiplication, S represents the scale of the vector, M x Represents the fine-grained interaction matrix of business contract content semantic understanding, Conv 3×3 (M x ) indicates that M x Perform convolutional coding based on a 3×3 convolution kernel, Ω represents the fine-grained interaction semantic matrix of business contract content semantic understanding, v kt represents the short-distance semantic understanding feature vector of the business contract content after information field mapping, v strepresents the feature vector of long-distance semantic understanding of business contract content after information field mapping, α and β represent weighted hyperparameters, and v c Represents the semantic understanding feature vector of the business contract content.

[0044] That is, in the technical solution of the present application, the feature scale adaptation of the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content is performed based on the target domain association matching. The process first maps the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content to the eigendecomposition space to obtain the short-distance semantic understanding eigenfeature vector of the business contract content and the long-distance semantic understanding eigenfeature vector of the business contract content. A person of ordinary skill in the art should know that the process of mapping to the eigendecomposition space is a combination of dimensionality reduction and feature extraction. In a specific example, the eigendecomposition space is found by the eigenvalue decomposition technique of principal component analysis (PCA), and the eigendecomposition space mapping is performed. The mapped eigenfeature vectors of short-distance semantic understanding of the business contract content and the eigenfeature vectors of long-distance semantic understanding of the business contract content not only represent the projection of the original feature vector in the new coordinate system, but also reflect the main direction of change of the data, which helps to avoid the "dimensionality disaster", that is, the problem that the sample density drops sharply as the number of features increases.

[0045] Next, after mapping the feature vector to the eigendecomposition space, it is necessary to perform target domain-oriented feature scaling adjustment on the eigenfeature vector of short-distance semantic understanding of business contract content and the eigenfeature vector of long-distance semantic understanding of business contract content to adjust the source domain feature vector to better match the data distribution of the target domain. This operation aims to capture and compensate for the systematic deviation between the two domains, so as to reduce the hidden disturbance of the class expression of the local feature distribution of the eigenfeature vector of short-distance semantic understanding of business contract content and the eigenfeature vector of long-distance semantic understanding of business contract content to the overall class expression. After modulation, the modulated eigenfeature vector of short-distance semantic understanding of business contract content and the modulated eigenfeature vector of long-distance semantic understanding of business contract content can retain the characteristics of the source domain while being more in line with the actual needs of the target domain, thereby enhancing the model's understanding and application of target domain data.

[0046] After completing the scale modulation of the feature vector, a fine-grained interactive semantic matrix of the business contract content semantic understanding is constructed between the intrinsic feature vector after the short-distance semantic understanding modulation of the business contract content and the intrinsic feature vector after the long-distance semantic understanding modulation of the business contract content. The fine-grained interactive semantic matrix of the business contract content semantic understanding is used to describe a more detailed relationship between feature vectors. The introduction of the fine-grained interactive semantic matrix of the business contract content semantic understanding enables the model to go beyond simple point-to-point comparison and consider more complex patterns and contextual information, thereby enhancing the model's sensitivity to subtle differences and improving the quality of feature representation. Finally, the intrinsic feature vector after the short-distance semantic understanding modulation of the business contract content and the intrinsic feature vector after the long-distance semantic understanding modulation of the business contract content are fused based on the fine-grained interactive semantic matrix of the business contract content semantic understanding to obtain the business contract content semantic understanding feature vector.

[0047] In step S140, based on the semantic understanding feature vector of the business contract content, the audit classification result of the business contract to be audited is obtained. Specifically, in the embodiment of the present application, the step S140 includes: passing the semantic understanding feature vector of the business contract content through a risk prompter based on a classifier to obtain an audit classification result, and the audit classification result is used to indicate whether the business contract to be audited has risks. It should be understood that in order to make a clear risk judgment based on the semantic features in the semantic understanding feature vector of the business contract content, it is necessary to input the semantic understanding feature vector of the business contract content into the risk prompter based on the classifier. A classifier is a machine learning model that can classify the input data into different categories according to the characteristics of the input data. Here, the role of the classifier is to judge the risk status of the business contract to be audited. The audit classification result intuitively indicates whether the business contract to be audited has risks, which is a clear judgment on the risk status of the contract. For the contract management party, the participating parties and other relevant entities, this is the most critical information, which can help them quickly know whether this contract may face risks such as breach of contract and clause disputes during the performance process, so as to decide whether further measures need to be taken, such as revising the clauses of the contract with risks and strengthening supervision. This will help achieve more scientific and effective contract risk management.

[0048] In summary, the data analysis method based on natural language processing technology in the embodiment of the present application is explained, which first obtains the business contract to be reviewed through the contract management system, then uses the data analysis technology based on natural language to perform content semantic analysis on the business contract to be reviewed, and automatically determines whether the business contract to be reviewed has risks based on the features obtained by the analysis. In this way, the problems of low efficiency, strong subjectivity and easy omission of risk points in traditional contract risk management can be effectively solved.

[0049] Figure 4 : is a system block diagram of a data analysis system based on natural language processing technology according to an embodiment of the present application. Figure 4 As shown, according to the data analysis system 100 based on natural language processing technology of the embodiment of the present application, it includes: a business contract data collection module 110 for review, which is used to obtain business contracts to be reviewed from the contract management system; a business contract data preliminary processing module 120 for reviewing, which is used to perform preliminary data processing on the business contract to be reviewed to obtain multiple business contract content word embedding vectors; a business contract data semantic analysis module 130 for reviewing, which is used to perform semantic analysis on the multiple business contract content word embedding vectors to obtain business contract content semantic understanding feature vectors; an audit result generation module 140, which is used to obtain the audit classification result of the business contract to be reviewed based on the business contract content semantic understanding feature vectors.

[0050] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the data analysis system 100 based on natural language processing technology have been described in detail above. Figures 1 to 3 The data analysis method based on natural language processing technology has been introduced in detail, and therefore, its repeated description will be omitted.

[0051] In summary, the data analysis system 100 based on natural language processing technology according to the embodiment of the present application is explained, which first obtains the business contract to be reviewed through the contract management system, then uses the data analysis technology based on natural language to perform content semantic analysis on the business contract to be reviewed, and automatically determines whether there is a risk in the business contract to be reviewed based on the features obtained by the analysis. In this way, the problems of low efficiency, strong subjectivity and easy omission of risk points in traditional contract risk management can be effectively solved.

[0052] As described above, the data analysis system 100 based on natural language processing technology according to the embodiment of the present application can be implemented in various wireless terminals, such as a server for data analysis based on natural language processing technology. In one example, the data analysis system 100 based on natural language processing technology according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the data analysis system 100 based on natural language processing technology can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the data analysis system 100 based on natural language processing technology can also be one of the many hardware modules of the wireless terminal.

[0053] Alternatively, in another example, the data analysis system 100 based on natural language processing technology and the wireless terminal may also be separate devices, and the data analysis system 100 based on natural language processing technology may be connected to the wireless terminal through a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

Claims

1. A data analysis method based on natural language processing technology, characterized in that: include: Obtain the business contracts to be reviewed from the contract management system; Performing preliminary data processing on the business contract to be reviewed to obtain a plurality of business contract content word embedding vectors; Performing semantic analysis on the multiple business contract content word embedding vectors to obtain a business contract content semantic understanding feature vector; Based on the semantic understanding feature vector of the business contract content, an audit classification result of the business contract to be audited is obtained.

2. The data analysis method based on natural language processing technology according to claim 1 is characterized in that: The business contract to be reviewed is preliminarily processed to obtain a plurality of business contract content word embedding vectors, including: Preprocessing the business contract to be reviewed to obtain a preprocessed business contract; The preprocessed business contract is segmented and passed through a word embedding layer to obtain the word embedding vectors of the multiple business contract contents.

3. The data analysis method based on natural language processing technology according to claim 2 is characterized in that: Performing semantic analysis on the multiple business contract content word embedding vectors to obtain a business contract content semantic understanding feature vector, including: Performing short-distance semantic analysis on the multiple business contract content word embedding vectors to obtain a business contract content short-distance semantic understanding feature vector; Performing long-distance semantic analysis on the multiple business contract content word embedding vectors to obtain a long-distance semantic understanding feature vector of the business contract content; The feature scale adaptation of the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content is performed based on target domain association matching to obtain the semantic understanding feature vector of the business contract content.

4. The data analysis method based on natural language processing technology according to claim 3 is characterized in that: Performing short-distance semantic analysis on the multiple business contract content word embedding vectors to obtain the business contract content short-distance semantic understanding feature vector, including: passing the multiple business contract content word embedding vectors through a first distance semantic capturer to obtain the business contract content short-distance semantic understanding feature vector.

5. The data analysis method based on natural language processing technology according to claim 4 is characterized in that: Performing long-distance semantic analysis on the multiple business contract content word embedding vectors to obtain the business contract content long-distance semantic understanding feature vector, including: passing the multiple business contract content word embedding vectors through a second distance semantic capturer to obtain the business contract content long-distance semantic understanding feature vector.

6. The data analysis method based on natural language processing technology according to claim 5 is characterized in that: The first distance semantic capturer is a recurrent neural network model, and the second distance semantic capturer is a bidirectional long short-term memory neural network model.

7. The data analysis method based on natural language processing technology according to claim 6 is characterized in that: The feature scale adaptation of the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content based on target domain association matching is performed to obtain the semantic understanding feature vector of the business contract content, including: Mapping the short-distance semantic understanding feature vector of the business contract content and the long-distance semantic understanding feature vector of the business contract content to an eigendecomposition space to obtain an eigenfeature vector of the short-distance semantic understanding of the business contract content and an eigenfeature vector of the long-distance semantic understanding of the business contract content; Performing target domain-oriented feature scale adjustment on the short-distance semantic understanding intrinsic feature vector of the business contract content and the long-distance semantic understanding intrinsic feature vector of the business contract content to obtain a modulated intrinsic feature vector of the short-distance semantic understanding of the business contract content and a modulated intrinsic feature vector of the long-distance semantic understanding of the business contract content; Constructing a fine-grained interactive semantic matrix of business contract content semantic understanding between the intrinsic feature vector after short-distance semantic understanding modulation of the business contract content and the intrinsic feature vector after long-distance semantic understanding modulation of the business contract content; Based on the fine-grained interactive semantic matrix of the business contract content semantic understanding, the modulated intrinsic feature vector of the business contract content short-distance semantic understanding and the modulated intrinsic feature vector of the business contract content long-distance semantic understanding are fused to obtain the business contract content semantic understanding feature vector.

8. The data analysis method based on natural language processing technology according to claim 7 is characterized in that: Based on the fine-grained interactive semantic matrix of the business contract content semantic understanding, the modulated intrinsic feature vector of the business contract content short-distance semantic understanding and the modulated intrinsic feature vector of the business contract content long-distance semantic understanding are fused to obtain the semantic understanding feature vector of the business contract content, including: Multiplying the fine-grained interactive semantic matrix of the business contract content semantic understanding by the modulated intrinsic feature vector of the business contract content short-distance semantic understanding to obtain a feature vector of the business contract content short-distance semantic understanding after information field mapping; Multiplying the fine-grained interactive semantic matrix of the business contract content semantic understanding by the modulated intrinsic feature vector of the business contract content long-distance semantic understanding to obtain a feature vector of the business contract content long-distance semantic understanding after information field mapping; The short-distance semantic understanding feature vector of the business contract content after the information field mapping and the long-distance semantic understanding feature vector of the business contract content after the information field mapping are weightedly fused to obtain the semantic understanding feature vector of the business contract content.

9. The data analysis method based on natural language processing technology according to claim 8, characterized in that: Based on the semantic understanding feature vector of the business contract content, an audit classification result of the business contract to be audited is obtained, including: passing the semantic understanding feature vector of the business contract content through a classifier-based risk indicator to obtain an audit classification result, and the audit classification result is used to indicate whether there is a risk in the business contract to be audited.

10. A data analysis system based on natural language processing technology, characterized in that: include: The data collection module for business contracts to be reviewed is used to obtain business contracts to be reviewed from the contract management system; A preliminary processing module for business contract data to be reviewed, used for preliminary processing of the data of the business contract to be reviewed to obtain a plurality of business contract content word embedding vectors; A semantic analysis module for the business contract data to be reviewed, used to perform semantic analysis on the multiple business contract content word embedding vectors to obtain a business contract content semantic understanding feature vector; The audit result generation module is used to obtain the audit classification result of the business contract to be audited based on the semantic understanding feature vector of the business contract content.

Images