An electronic contract management method and system based on deep learning model

Abstract

The present application provides an electronic contract management method and system based on a deep learning model, which relates to the field of computer data processing technology. The method includes: identifying the contract clause classification corresponding to each contract clause in the target electronic contract to be managed; for each corresponding contract clause, using a clause risk information identification model that matches the contract clause classification of the contract clause to perform risk identification to extract key risk information in the contract clause, determining the clause risk score corresponding to the contract clause based on a risk scoring model, and generating an audit interaction control based on the contract clause and the key risk information in the contract clause when it is determined that the clause risk score exceeds a preset scoring threshold; when all the audit interaction controls are confirmed, the target electronic contract is classified and stored in an electronic contract archive database according to the archive tag. Thus, the review process is simplified and the intelligent level of contract management is greatly improved.

Description

Technical Field

[0001] The present application relates to the field of computer data processing technology, and in particular to an electronic contract management method and system based on a deep learning model. Background Art

[0002] In today's business environment, electronic contracts have become an important tool for businesses and individuals to conduct transactions and cooperation. Compared with traditional paper contracts, electronic contracts have the advantages of easy signing, fast transmission, and simple storage. Therefore, more and more companies choose electronic contracts to replace traditional paper contracts. However, with the popularization of electronic contracts, new challenges have also emerged in contract management.

[0003] Currently, there are a variety of electronic contract management software and services on the market. These tools mainly provide contract creation, signing, storage and retrieval functions. However, in actual applications, existing electronic contract management systems usually rely on manual review and confirmation of contract content, which is not only time-consuming and labor-intensive, but also prone to human errors. Especially when dealing with a large number of contracts, the efficiency and accuracy of manual review are difficult to guarantee.

[0004] In manual contract review, the professional level and experience of the reviewer will directly affect the fluctuation of the review results. Different reviewers may have different understandings and judgments on the same contract terms, resulting in inconsistency and lack of accuracy in the review results. In addition, manual review is difficult to cover all the details and potential risks in the contract, and it is easy to miss important terms or risk points.

[0005] In response to the above problems, the industry has not yet proposed a better technical solution. Summary of the invention

[0006] The present application provides an electronic contract management method, system, storage medium, computer program product and electronic device based on a deep learning model, which is used to at least solve the current problems of low efficiency and easy omission of clause risk points in manual review of electronic contracts.

[0007] In a first aspect, an embodiment of the present application provides an electronic contract management method based on a deep learning model, comprising: identifying a contract clause classification corresponding to each contract clause in a target electronic contract to be managed; the contract clause classification includes at least one of the following: payment clauses, breach of contract clauses, confidentiality clauses, term clauses, dispute resolution clauses, contract termination clauses and liability clauses; for each of the contract clauses, performing the following operations respectively: using a clause risk information identification model that matches the contract clause classification of the contract clause to perform risk identification to extract key risk information in the contract clause, inputting the key risk information and the contract clause classification that matches the contract clause into a risk scoring model to determine a clause risk score corresponding to the contract clause, and when it is determined that the clause risk score exceeds a preset scoring threshold, generating an audit interaction control based on the contract clause and the key risk information in the contract clause; the clause risk information identification model and the risk scoring model adopt a deep learning model; when all of the audit interaction controls in the target electronic contract detect a confirmation interaction operation, determining the archive tag corresponding to the target electronic contract, and storing the target electronic contract classification in an electronic contract archive database according to the archive tag.

[0008] In a second aspect, an embodiment of the present application provides an electronic contract management system based on a deep learning model, including: a clause classification identification unit, used to identify the contract clause classification corresponding to each contract clause in the target electronic contract to be managed; the contract clause classification includes at least one of the following: payment clauses, breach of contract clauses, confidentiality clauses, term clauses, dispute resolution clauses, contract termination clauses and liability clauses; a key information identification unit, used to perform risk identification for each of the contract clauses using a clause risk information identification model that matches the contract clause classification of the contract clause to extract key risk information from the contract clause; a clause risk scoring unit, used to compare the key risk information with the contract clause classification; The contract clause classification that matches the contract clause is input into the risk scoring model to determine the clause risk score corresponding to the contract clause; the interactive control generation unit is used to generate an audit interactive control based on the contract clause and the key risk information in the contract clause when it is determined that the clause risk score exceeds a preset scoring threshold; the clause risk information identification model and the risk scoring model adopt a deep learning model; the electronic contract archiving unit is used to determine the archiving label corresponding to the target electronic contract when all the audit interactive controls in the target electronic contract detect a confirmation interaction operation, and classify the target electronic contract and store it in the electronic contract archive database according to the archiving label.

[0009] According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the steps of the electronic contract management method based on a deep learning model of any embodiment of the present application.

[0010] In a fourth aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the electronic contract management method based on a deep learning model of any embodiment of the present application are implemented.

[0011] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the electronic contract management method based on a deep learning model of any embodiment of the present application.

[0012] The electronic contract management method based on the deep learning model provided by this application can produce at least the following technical effects:

[0013] (1) By automatically identifying and classifying contract terms, and implementing automated risk information identification and risk scoring models for each term, it is possible to more accurately extract key risk information in contract terms and perform risk scoring. Since the deep learning model can learn and extract risk features from a large amount of historical contract data, it can automatically identify potential risk points in contract terms and remind reviewers to pay attention to high-risk terms and risk points, ensuring that important terms and risk points are not missed, significantly reducing the workload and time of manual review, improving review efficiency, and ensuring the comprehensiveness and effectiveness of risk management.

[0014] (2) Dynamically evaluate each contract clause through the risk scoring model, and generate corresponding audit interaction controls based on the situation where the risk score of the clause exceeds the preset threshold. The system can adjust the risk assessment in real time according to the changes in the contract content, providing flexible and timely risk management. After all audit interaction controls are confirmed, the system will automatically determine the contract filing label and store the contract classification in the electronic contract archive database, reducing the workload of manual classification and storage, making contract management more convenient and systematic.

[0015] (3) After all audit interaction controls in the target electronic contract have been confirmed, archive tags are generated based on the contract terms and risk information, and the contracts are classified and stored in the electronic contract archive database to ensure the standardization and orderliness of contract archiving. In this way, the automatic generation and classification storage functions of archive tags are realized, making contract retrieval more efficient and accurate, and facilitating enterprises to quickly find and use contract information in subsequent management.

[0016] Through this technical solution, the entire contract management process has been highly automated, from clause classification, risk identification, risk scoring to the generation of audit controls and archiving storage, integrating natural language technology and deep learning models for automated processing. Reviewers only need to focus on the key risk information in each contract clause, which simplifies the review process and greatly improves the level of intelligence in contract management. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A flowchart showing an example of an electronic contract management method based on a deep learning model according to an embodiment of the present application is shown;

[0019] Figure 2 An operation flow chart showing an example of identifying the classification of contract terms according to an embodiment of the present application is shown;

[0020] Figure 3 A schematic diagram showing a structural connection of an example of a clause risk information identification model according to an embodiment of the present application is shown;

[0021] Figure 4 A schematic diagram showing a structural connection of an example of a risk scoring model according to an embodiment of the present application is shown;

[0022] Figure 5 A structural block diagram of an example of an electronic contract management system based on a deep learning model according to an embodiment of the present application is shown;

[0023] Figure 6 It is a schematic structural diagram of an embodiment of an electronic device of the present application. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0025] In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.

[0026] Figure 1 A flowchart of an example of an electronic contract management method based on a deep learning model according to an embodiment of the present application is shown.

[0027] Regarding the executor of the method of the embodiment of the present application, it can be any controller or processor with computing or processing capabilities, which significantly improves the efficiency, accuracy and intelligence level of contract management by automating contract clause classification, risk identification, risk scoring, review interaction control generation and contract archiving.

[0028] In some examples, it can be an electronic contract management platform, and can be integrated and configured in an electronic device or terminal through software, hardware, or a combination of software and hardware, and the type of terminal or electronic device can be diverse, such as a mobile phone, tablet computer, or desktop computer, etc.

[0029] like Figure 1 As shown, in step S110, the contract clause classification corresponding to each contract clause in the target electronic contract to be managed is identified.

[0030] More specifically, the electronic contract text can be preprocessed, including format removal, segmentation, sentence separation, etc., and then the standardized data can be classified. In an example of an embodiment of the present application, a trained deep learning model (such as BERT, GPT, etc.) is used to classify the preprocessed contract text to identify the corresponding contract clause classification. The contract clause classification includes at least one of the following: payment clause, breach of contract clause, confidentiality clause, term clause, dispute resolution clause, contract termination clause, and liability clause.

[0031] Figure 2 FIG. 1 is a flowchart showing an example of identifying the classification of contract terms according to an embodiment of the present application. Figure 2As shown, in step S210, for each contract clause in the target electronic contract, a plurality of preliminary keywords are extracted from the clause text corresponding to the contract clause using the TF-IDF algorithm. Here, the words are sorted according to the TF-IDF value, the first several keywords are selected as preliminary keywords, and the preliminary keyword list of each contract clause is output. Therefore, the TF-IDF algorithm is simple to calculate and suitable for large-scale data processing. It can quickly extract preliminary keywords in the contract clauses and filter out important words in the contract clauses by the TF-IDF value. In step S220, each preliminary keyword is scored based on the RAKE algorithm, and the target keywords with high scores are retained. The RAKE (Rapid Automatic Keyword Extraction) algorithm is a rapid automatic keyword extraction algorithm that scores and extracts keywords by analyzing the co-occurrence relationship and the position of the words. Then, the keywords with the highest scores are selected as the target keywords, which can identify the phrases and compound words in the contract clauses and improve the accuracy of keyword extraction. In step S230, a preset keyword table is matched with each target keyword, and the contract clause classification corresponding to the contract clause is determined according to the matching result. The keyword table contains multiple calibrated keywords and corresponding calibrated contract clause classifications. In this way, the classification of the contract clause is determined according to the matching result by matching the target keyword extracted by the RAKE algorithm with the preset keyword table. If a clause contains keywords of multiple classifications, priority rules or multiple classifications can be set. For example, a clause containing keywords such as "legal liability" and "litigation" is classified as "legal risk". Therefore, by matching the preset keyword table with the target keyword, the contract clause can be accurately identified and classified. In addition, the keyword table can be expanded and optimized according to actual needs to adapt to new contract clauses and classification requirements.

[0032] In step S120, for each corresponding contract clause, a clause risk information identification model that matches the contract clause classification of the contract clause is used to perform risk identification to extract key risk information in the contract clause.

[0033] More specifically, a specially trained deep learning model is used to identify and extract risk information for each type of contract clause. The model extracts potential key risk information by analyzing the content of the clause, such as the risk of delayed payment in the payment clause, the liability for breach of contract in the breach of contract clause, etc. The model is trained based on a large amount of labeled data to ensure accurate identification of various types of risk information. In this way, key risk information in the contract can be extracted quickly and accurately, reducing errors in manual identification, covering all major risk types, and ensuring a comprehensive risk assessment of the contract.

[0034] It should be noted that different types of contract clauses (such as payment clauses, breach of contract clauses, confidentiality clauses, etc.) have their own unique content and structure. Since these clauses involve different topics and risk points, by setting up a special model, the model can be trained more specifically to more accurately identify the key risk information in such clauses. In addition, corresponding clause risk information identification models are set up for each contract clause classification. The electronic contract management platform needs to install multiple customized clause risk information identification models. Therefore, it is recommended to use a lightweight model for the clause risk information identification model.

[0035] In some examples of the embodiments of the present application, the key risk information includes the text of the risk clause and the corresponding risk category, and the risk category includes any one of the following: legal risk type, financial risk type, compliance risk type, reputation risk type, and technical risk type. Thus, not only the key risk text can be identified and marked, but also the corresponding risk category can be identified, and the model can provide a comprehensive risk assessment. In addition, different risk categories have corresponding characteristics, and through the accurate classification of risk categories by the model, more targeted countermeasures can be taken.

[0036] In step S130, the key risk information and the contract clause classification matching the contract clause are input into the risk scoring model to determine the clause risk score corresponding to the contract clause.

[0037] Here, the risk scoring model can use deep learning technology to score the input risk information and corresponding clause classification according to the preset scoring criteria and historical data, and then output the risk score of each clause. In this way, a quantitative risk score is provided for each contract clause, which is convenient for intuitive comparison and evaluation. The model can continuously update and optimize the scoring criteria according to new contract data and risk information, and improve the accuracy and timeliness of the scoring.

[0038] In step S140, when it is determined that the clause risk score exceeds a preset score threshold, an audit interaction control is generated based on the contract clause and key risk information in the contract clause.

[0039] Specifically, according to the enterprise risk management strategy, the risk score thresholds of various clauses are preset. When the risk score of a clause exceeds the threshold, the system automatically generates the corresponding audit interactive control. Exemplarily, the interactive control content includes the specific content of the clause, key risk information, recommended response measures, etc., which can warn the auditors to pay attention and avoid ignoring important risks. Therefore, through the interactive control, intuitive risk information and suggestions are provided, which facilitates auditors to make decisions quickly.

[0040] In step S150, when all the review interaction controls in the target electronic contract detect the confirmation interaction operation, the archiving tag corresponding to the target electronic contract is determined, and the target electronic contract is classified and stored in the electronic contract archive database according to the archiving tag.

[0041] In some implementations, the electronic contract management platform monitors the confirmation of each audit interaction control to ensure that all high-risk clauses are audited and confirmed, and determines the contract filing label based on the audit results and contract content, such as a label determined based on business attributes or audit results. Contracts are classified and stored in the electronic contract archive database according to the filing labels to facilitate subsequent retrieval and management, thereby automatically filing and classifying storage, making contract management more systematic and orderly, and improving management efficiency.

[0042] Through the embodiments of the present application, the full process of electronic contract management is automated and intelligent, from contract clause classification, risk identification, risk scoring to generation of audit controls and archiving storage, which comprehensively improves the efficiency, accuracy and consistency of contract management, ensures the accurate identification and assessment of contract clauses and risk information, greatly reduces manual intervention and human errors, and improves the risk control and decision-making support capabilities of enterprises in contract management.

[0043] Figure 3 A structural connection diagram of an example of a clause risk information identification model according to an embodiment of the present application is shown.

[0044] like Figure 3 As shown, the clause risk information identification model 300 includes a cascaded input layer 310, an embedding layer 320, an encoding layer 330, a knowledge graph embedding layer 340, an attention mechanism layer 350 and an output layer 360.

[0045] In some embodiments, a knowledge graph is a graph structure used to represent entities and their relationships. In contract management, key information in contract terms can be mapped to the knowledge graph to facilitate a deeper semantic understanding of the model. More specifically, nodes represent key entities in contract terms, such as company name, date, amount, location, etc., so each node contains rich semantic information, which helps the model understand the specific content of the terms. In addition, edges represent the relationship between entities, such as payment relationship, breach of contract liability, contract term, etc., so the type and direction of the edge can help the model understand the semantic association between different entities.

[0046] Here, the knowledge graph enables the model to identify more potential risk information through rich node and edge relationships. Especially when dealing with complex contract terms across fields, the knowledge graph provides extensive background information and enhances the model's recognition ability. For example, in payment terms, in addition to the direct payment amount and date, the model can also identify payment-related conditions and restrictions. For example, in breach of contract terms, the model can not only identify breach of contract, but also find related legal consequences and liability terms through the knowledge graph.

[0047] The input layer 310 is used to determine the word sequence corresponding to the contract clause:

[0048] T={w1,w2,…,w n}, formula (1)

[0049] Where T represents the input contract terms text, w i Represents the i-th word in the clause text.

[0050] In some embodiments, the contract clause text is preprocessed, including word segmentation, stop word removal, lemma restoration, etc. In addition, the preprocessed text is converted into a word sequence, and a unique index is assigned to each word. Then, the word sequence is padded or truncated to a fixed length. In this way, the input data format is unified, which is convenient for subsequent processing.

[0051] The embedding layer 320 is used to convert the word sequence into a vector representation of fixed dimension:

[0052] e i =E(w i ), formula (2)

[0053] In the formula, e i represents the embedding vector of the i-th word, and E represents the pre-trained word embedding matrix.

[0054] In some embodiments, a pre-trained word embedding matrix (such as Word2Vec, GloVe or BERT) is used to initialize word embedding to convert each word in a word sequence into a corresponding embedding vector. Thus, through the pre-trained word embedding, the semantic information of the word can be captured, and the high-dimensional text data can be converted into a low-dimensional vector representation for subsequent processing.

[0055] The encoding layer 330 uses the Transformer model to encode the input sequence to obtain the corresponding text encoding vector:

[0056] h i =f(e1,e2,...,e n ), formula (3)

[0057] Where f represents the Transformer encoding function, h i Represents the context vector representation of the i-th word.

[0058] Here, the embedded representation is input into the Transformer encoder to perform multi-head self-attention calculation and feedforward neural network calculation. As a result, the Transformer model can capture long-distance dependencies in text sequences and improve text comprehension capabilities.

[0059] The knowledge graph embedding layer 340 combines the text encoding vector with the nodes and edges in the knowledge graph to generate a more semantically understandable semantic embedding representation:

[0060] g i =GE(G,h i ), formula (4)

[0061] Among them, G represents the knowledge graph, GE() represents the graph embedding operation, and g i Represents the semantic representation vector of the i-th word after combining text encoding and knowledge graph embedding.

[0062] Here, the nodes in the knowledge graph represent key entities in the contract terms, and the key entities include at least one of the following: company name, date, amount, and location. The edges in the knowledge graph represent entity relationships in the contract terms, and the entity relationships include at least one of the following: payment relationship, breach of contract liability, and contract term.

[0063] In an embodiment of the present application, by constructing a knowledge graph of contract terms, the key entities and relationships in the contract are systematically represented, and the model can better understand the semantic information of the contract terms. The knowledge graph embedding layer combines the text vector with the knowledge graph, so that the model can capture deeper semantic relationships when dealing with complex terms. As a result, risk points in the terms are more accurately identified. For example, in terms involving complex legal relationships, the model can better understand the relevant laws and legal responsibilities through the knowledge graph. In particular, for the different meanings of the same legal term in different contexts, the model can distinguish them based on the knowledge graph, which can improve the accuracy of risk identification.

[0064] The attention mechanism layer 350 is used to process the semantic embedding representation to calculate the importance of each word in the overall clause text, thereby generating an attention context vector:

[0065]

[0066] In the formula, g j and g k They represent the semantic representation vectors of the jth and kth words after combining text encoding and knowledge graph embedding, gj Used to calculate the similarity with the i-th word, g k Used to calculate the normalized attention weight; α ij represents the attention weight of the jth word to the ith word; c i Represents the weighted context vector of the i-th word.

[0067] The importance of each word in the overall clause text is calculated through the attention mechanism layer 350, thereby generating an attention context vector. Specifically, the similarity score of each word with other words is calculated, normalized by calculating the attention weight, and the semantic representation vector is weighted and summed according to the attention weight to generate the attention context vector. Thus, through the attention mechanism, the model can identify and focus on key information in the contract terms, improve the accuracy of risk identification, dynamically assign weights according to the context, and better understand and process important information in the text.

[0068] The output layer 360 is used to process the attention context vector to generate prediction results of key risk information:

[0069] Z = tanh(W FC c+b FC ), formula (7)

[0070] Y risk =softmax(Z), formula (8)

[0071] Where c represents the attention context vector calculated by the attention mechanism layer, W FC and b FC They represent the weight matrix and bias vector of the fully connected layer, tanh represents the hyperbolic tangent activation function, Z represents the intermediate feature representation of the output of the fully connected layer; softmax represents the softmax activation function, Y risk Represents the key risk information corresponding to the predicted key risk information m.

[0072] In an embodiment of the present application, the attention context vector is processed based on the output layer to generate a prediction result of key risk information. Through processing by the fully connected layer and the softmax layer, accurate prediction results of key risk information are generated to facilitate further risk assessment and decision-making.

[0073] Through the embodiments of this application, combined with Transformer encoding and knowledge graph embedding, the model can fully understand the semantic information in the contract terms and improve the accuracy of risk identification. Through the attention mechanism, the model can identify and focus on the key information in the contract terms and improve the semantic understanding ability. Combining knowledge graph and text encoding, a dynamic fusion semantic representation is generated, which is suitable for risk identification and management of large-scale electronic contracts.

[0074] During the practice of this application, combined with the characteristics of the electronic contract management scenario, we also made a series of enhancement designs to the encoding layer and the knowledge graph embedding layer to improve their advantages in this scenario.

[0075] In some examples of the embodiments of the present application, the encoding layer adopts a deep Transformer encoder based on a multi-head self-attention mechanism, which includes a cascaded multi-head self-attention mechanism layer and a multi-layer Transformer encoder, which can significantly improve the model's semantic understanding and risk identification capabilities for contract terms. Specifically, based on the multi-head self-attention mechanism layer, the input embedding vector sequence is processed by the multi-head self-attention mechanism. Furthermore, the vector sequence output by the multi-head self-attention mechanism layer is processed by a multi-layer Transformer encoder to obtain a corresponding text encoding vector.

[0076] Through the embodiments of the present application, the multi-head self-attention mechanism can simultaneously focus on different parts of the contract terms text, capture more semantic information, and enable the model to more accurately understand the meaning of the contract terms. Specifically, the multi-head self-attention mechanism uses multiple attention heads to focus on different contextual information, enabling the model to capture fine-grained semantic features in the contract terms. For example, when analyzing "payment terms", the model can not only identify the payment amount and date, but also understand the semantic information related to the payment terms. In addition, the multi-head mechanism can reduce the information confusion that may be caused by a single attention head, allowing the model to understand the same clause from multiple perspectives and improve the accuracy of semantic understanding. In addition, the self-attention mechanism can directly model the dependencies between distant words in the contract terms, overcoming the disadvantages of traditional RNNs in processing long texts. For example, in complex "breach of contract terms", the model can accurately identify distantly related breach of contract liability and compensation terms.

[0077] In some examples of the embodiments of the present application, the knowledge graph embedding layer embeds nodes in the knowledge graph using a graph neural network (GNN) and fuses them with text vector representations, which can significantly improve the model's semantic understanding of contract terms and risk identification capabilities.

[0078] In the embodiment of the present application, the knowledge graph embedding layer maps the key entities and relationships in the contract terms to the knowledge graph, and embeds the nodes through the graph neural network, so that the model can capture deeper semantic information and contextual relationships. Therefore, through the knowledge graph, the model can understand the complex relationships and semantic associations between entities in the contract terms, and based on the graph neural network through a multi-layer message passing mechanism, extract and fuse the semantic information of the nodes layer by layer, so that the embedded representation of each node is more comprehensive and rich. For example, the model can accurately understand the details of the legal responsibilities and penalty for breach of contract involved in the "breach of contract clause".

[0079] Specifically, the knowledge graph embedding layer is used to embed nodes in the knowledge graph using a graph neural network:

[0080] v p =GNN(G,v p ), formula (9)

[0081] In the formula, v p is the embedded representation of node p in the knowledge graph, and G is the knowledge graph.

[0082] In graph neural networks, the embedding representation of a node is updated via a message passing mechanism:

[0083]

[0084] Where N(p) represents the set of neighbor nodes of node p, σ is the activation function, and W (k) and b (k) are the weight matrix and bias vector of the kth layer respectively; represents the embedding vector of node q at layer k, Represents the embedding vector of node p in the k+1th layer.

[0085] The text vector representation is fused with the node embedding of the knowledge graph to generate the final embedding representation:

[0086] g i =concat(h i ,v p ), formula (11)

[0087] In the formula, concat represents the feature concatenation operation.

[0088] In an embodiment of the present application, the knowledge graph embedding layer fuses the text vector with the node embedding of the knowledge graph, so that the model can combine the information in the text and the graph during the risk identification process, thereby improving the comprehensiveness and accuracy of risk identification. By splicing the text vector with the node embedding of the knowledge graph, the model can use the information in the text and the graph for risk identification at the same time. For example, in the "confidentiality clause", the model not only identifies the sensitive information in the clause text, but also combines the legal clauses in the knowledge graph for risk assessment. Therefore, through the embedded representation of the graph neural network, the implicit risks and nuances in the contract terms can be captured, making risk identification more accurate. For example, the model can identify hidden clauses in the "dispute resolution clause" that are unfavorable to the enterprise.

[0089] Figure 4 A schematic diagram of the structural connection of an example of a risk scoring model according to an embodiment of the present application is shown.

[0090] like Figure 4 As shown, the risk scoring model 400 includes an input layer 410, a bidirectional LSTM layer 420, a multi-scale convolutional neural network layer 430 and an output layer 440.

[0091] The input layer 410 is used to determine the fused input features corresponding to the key risk information and contract clause classification:

[0092] d m =concat(d m_risk ,d m_clause ), formula (12)

[0093] Where, d m_risk represents the embedding representation of key risk information m, d m_clause The embedding representation of the corresponding contract clause classification, d m Represents the corresponding fused input features.

[0094] The bidirectional LSTM layer 420 is used to capture the contextual information of key risk information and contract terms and generate forward and backward hidden states:

[0095]

[0096] In the formula, and They respectively represent the forward LSTM hidden state and backward LSTM hidden state corresponding to the key risk information m.

[0097] Bidirectional LSTM can capture forward and backward contextual information in the input sequence, improving the model's ability to understand contract terms and key risk information. Through BiLSTM's bidirectional processing, the model can simultaneously capture the contextual dependencies between contract terms and key risk information, ensuring a comprehensive understanding of complex contract terms. For example, when analyzing breach of contract terms, the model can take into account the content before and after the terms to generate a more accurate risk score. BiLSTM can understand the semantics of contract terms and key risk information at a deeper level, especially when processing long texts and complex sentences.

[0098] The multi-scale convolutional neural network layer 430 uses convolution kernels of different sizes for feature extraction to capture multi-scale feature information:

[0099]

[0100] In the formula, is the output feature representation of the key risk information m in the kth convolution kernel, r m Represents the multi-scale feature information corresponding to the key risk information m; and They represent the weight matrix and bias vector of the kth convolution kernel respectively, * represents the convolution operation, and ReLU represents the RELU activation function.

[0101] Multi-scale convolutional neural networks extract multi-scale features through convolution kernels of different sizes, capturing local patterns and high-order features of input features. Specifically, convolution kernels of different sizes can capture information of different granularities, such as phrase-level and sentence-level features in contract terms, making the model's understanding of the contract more detailed and comprehensive. As a result, multi-scale convolutional neural networks can effectively extract key information in contract terms and improve the model's ability to identify potential risks. For example, for complex payment terms, the model can accurately identify the core payment terms and related risk factors.

[0102] The structure of the output layer 440 is as follows:

[0103] f m =ReLU(W f r m +b f ), formula (17)

[0104] z m =tanh(W z f m +b z ), formula (18)

[0105] y m =softmax(W y z m +by ), formula (19)

[0106] In the formula, f m represents the output feature of the fully connected layer, W f , W z and W y Represents the weight matrices of different layers of the output layer, b f , b z and b y Respectively represent the bias items of different layers of the output layer; z m represents the intermediate representation vector; y m Represents the clause risk score corresponding to the predicted key risk information m.

[0107] Therefore, in the output layer, the functions of the fully connected layer and the output layer are merged, and a more accurate risk score is generated through multi-layer full connection and weighting mechanism.

[0108] Through the risk scoring model of the embodiment of the present application, the multi-scale convolutional neural network and the bidirectional LSTM are used for feature fusion, and the multi-scale convolutional neural network with parallel computing is adopted, so that the model is more efficient when processing large-scale data. Through the bidirectional LSTM and multi-layer feature extraction, the model shows stronger robustness and stability when facing uncertain and complex contract terms, reduces the misjudgment rate, and realizes a comprehensive risk assessment of the contract terms.

[0109] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of actions combined, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0110] Figure 5 A structural block diagram of an example of an electronic contract management system based on a deep learning model according to an embodiment of the present application is shown.

[0111] like Figure 5 As shown, the electronic contract management system 500 based on the deep learning model includes a clause classification identification unit 510, a key information identification unit 520, a clause risk scoring unit 530, an interactive control generation unit 540 and an electronic contract archiving unit 550.

[0112] The clause classification identification unit 510 is used to identify the contract clause classification corresponding to each contract clause in the target electronic contract to be managed; the contract clause classification includes at least one of the following: payment clause, breach of contract clause, confidentiality clause, term clause, dispute resolution clause, contract termination clause and liability clause.

[0113] The key information identification unit 520 is used to perform risk identification for each of the contract clauses using a clause risk information identification model that matches the contract clause classification of the contract clause, so as to extract key risk information in the contract clause.

[0114] The clause risk scoring unit 530 is used to input the key risk information and the contract clause classification matching the contract clause into the risk scoring model to determine the clause risk score corresponding to the contract clause.

[0115] The interactive control generation unit 540 is used to generate an audit interactive control based on the contract clause and the key risk information in the contract clause when it is determined that the clause risk score exceeds a preset score threshold; the clause risk information identification model and the risk scoring model adopt a deep learning model.

[0116] The electronic contract archiving unit 550 is used to determine the archiving tag corresponding to the target electronic contract when all the review interaction controls in the target electronic contract detect the confirmation interaction operation, and classify and store the target electronic contract in the electronic contract archive database according to the archiving tag.

[0117] In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, which stores one or more programs including execution instructions, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the steps of any of the above-mentioned electronic contract management methods based on deep learning models in the present application.

[0118] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer performs any step of the above-mentioned electronic contract management method based on a deep learning model.

[0119] In some embodiments, an embodiment of the present application also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the electronic contract management method based on a deep learning model.

[0120] Figure 6 is a schematic diagram of the hardware structure of an electronic device for executing an electronic contract management method based on a deep learning model provided by another embodiment of the present application, such as Figure 6 As shown, the device includes:

[0121] One or more processors 610 and memory 620, Figure 6 A processor 610 is taken as an example.

[0122] The device for executing the electronic contract management method based on the deep learning model may further include: an input device 630 and an output device 640 .

[0123] The processor 610, the memory 620, the input device 630 and the output device 640 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0124] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the electronic contract management method based on the deep learning model in the embodiment of the present application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 620, that is, the electronic contract management method based on the deep learning model in the above method embodiment is implemented.

[0125] The memory 620 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 620 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include a memory remotely arranged relative to the processor 610, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0126] The input device 630 can receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 640 can include a display device such as a display screen.

[0127] The one or more modules are stored in the memory 620, and when executed by the one or more processors 610, the electronic contract management method based on the deep learning model in any of the above method embodiments is executed.

[0128] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0129] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:

[0130] (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0131] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.

[0132] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0133] (4) Other onboard electronic devices with data interaction functions, such as on-board devices installed in vehicles.

[0134] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0135] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An electronic contract management method based on a deep learning model, comprising: Identify the contract clause classification corresponding to each contract clause in the target electronic contract to be managed; the contract clause classification includes at least one of the following: payment clause, breach of contract clause, confidentiality clause, term clause, dispute resolution clause, contract termination clause and liability clause; For each of the contract terms, perform the following operations: Using a clause risk information identification model that matches the contract clause classification of the contract clause to perform risk identification to extract key risk information in the contract clause, Inputting the key risk information and the contract clause classification matching the contract clause into the risk scoring model to determine the clause risk score corresponding to the contract clause, When it is determined that the risk score of the clause exceeds a preset score threshold, an audit interaction control is generated according to the contract clause and the key risk information in the contract clause; the clause risk information identification model and the risk scoring model adopt a deep learning model; In the case where all the review interaction controls in the target electronic contract detect a confirmation interaction operation, determining an archive tag corresponding to the target electronic contract, and classifying and storing the target electronic contract in an electronic contract archive database according to the archive tag; The risk scoring model includes an input layer A, a bidirectional LSTM layer, a multi-scale convolutional neural network layer and an output layer B; The input layer A is used to determine the fusion input features corresponding to the key risk information and contract clause classification: ， In the formula, Indicates key risk information m The embedding representation of Embedding representation representing the corresponding contract clause classification, Represents the corresponding fusion input features, and concat represents the feature concatenation operation; The bidirectional LSTM layer is used to capture the contextual information of key risk information and contract terms and generate forward and backward hidden states: ， ， In the formula, and Represents key risk information m The corresponding forward LSTM hidden state and backward LSTM hidden state; The multi-scale convolutional neural network layer uses convolution kernels of different sizes for feature extraction to capture multi-scale feature information: ， ， In the formula, Corresponding key risk information m In the The output feature representation of the convolution kernel is: Indicates key risk information m The corresponding multi-scale feature information; and Respectively represent The weight matrix and bias vector of the convolution kernel, represents the convolution operation, Represents the RELU activation function; The structure of the output layer B is as follows: ， ， ， In the formula, represents the output feature of the fully connected layer, , and Represent the weight matrices of different layers of the output layer B, , and Respectively represent the bias items of different layers of the output layer B; represents the intermediate representation vector, represents the hyperbolic tangent activation function; represents the softmax activation function, Indicates the predicted key risk information m The corresponding clause risk score.

2. The method according to claim 1, wherein the clause risk information identification model comprises a cascaded input layer C, an embedding layer, an encoding layer, a knowledge graph embedding layer, an attention mechanism layer and an output layer D; The input layer C is used to determine the word sequence corresponding to the contract terms: ， In the formula, Indicates the text of the contract terms entered. Indicates the first words; The embedding layer is used to convert the word sequence into a vector representation of fixed dimension: ， In the formula, Indicates The embedding vectors of words, Represents the pre-trained word embedding matrix; The encoding layer uses the Transformer model to encode the input sequence to obtain the corresponding text encoding vector: ， In the formula, represents the Transformer encoding function, Indicates The context vector representation of each word; The knowledge graph embedding layer combines the text encoding vector with the nodes and edges in the knowledge graph to generate a more semantically understandable semantic embedding representation: ， in, represents the knowledge graph, represents the graph embedding operation, Indicates The semantic representation vector of the combined text encoding and knowledge graph embedding of the words; the nodes in the knowledge graph represent the key entities in the contract terms, and the key entities include at least one of the following: company name, date, amount and location; the edges in the knowledge graph represent the entity relationships in the contract terms, and the entity relationships include at least one of the following: payment relationship, breach of contract liability and contract term; The attention mechanism layer is used to process the semantic embedding representation to calculate the importance of each word in the overall clause text, thereby generating an attention context vector: ， ， In the formula, and Respectively represent and The semantic representation vector of a word after combining text encoding and knowledge graph embedding, For calculation and The similarity of words, Used for calculation of normalized attention weights; Indicates Word pair The attention weight of each word; Indicates The weighted context vector of each word; The output layer D is used to process the attention context vector to generate a prediction result of key risk information: ， ， In the formula, represents the attention context vector calculated by the attention mechanism layer, and Represent the weight matrix and bias vector of the fully connected layer respectively, Represents the intermediate feature representation of the fully connected layer output; Represents the predicted key risk information.

3. The method according to claim 2, wherein: The encoding layer adopts a deep Transformer encoder based on a multi-head self-attention mechanism, which includes a cascaded multi-head self-attention mechanism layer and a multi-layer Transformer encoder; Based on the multi-head self-attention mechanism layer, the input embedding vector sequence is processed by the multi-head self-attention mechanism; The vector sequence output by the multi-head self-attention mechanism layer is processed by the multi-layer Transformer encoder to obtain a corresponding text encoding vector.

4. The method according to claim 2, wherein: The knowledge graph embedding layer is used to embed nodes in the knowledge graph using a graph neural network: ， In the formula, Is a node in the knowledge graph The embedding representation of It is a knowledge graph; In graph neural networks, the embedding representation of a node is updated through a message passing mechanism: ， In the formula, Representation Node The set of neighbor nodes of is the activation function, and They are The weight matrix and bias vector of the layer; Representation Node In the The embedding vector of the layer, Representation Node In the The embedding vector of the layer; The text vector representation is fused with the node embedding of the knowledge graph to generate the final embedding representation: 。 5. The method according to claim 1, wherein: The identifying of the contract clause classification corresponding to each contract clause in the target electronic contract to be managed includes: For each contract clause in the target electronic contract, Extract multiple preliminary keywords from the clause text corresponding to the contract clause using the TF-IDF algorithm; Scoring each of the preliminary keywords based on the RAKE algorithm, and retaining target keywords with scores above a certain threshold; Based on a preset keyword table, matching is performed with each of the target keywords, and the contract clause classification corresponding to the contract clause is determined according to the matching result; the keyword table includes multiple calibrated keywords and corresponding calibrated contract clause classifications.

6. The method according to any one of claims 1 to 5, wherein: The key risk information includes the text of risk clauses and corresponding risk categories; the risk categories include any one of the following: legal risk type, financial risk type, compliance risk type, reputation risk type and technical risk type.

7. An electronic contract management system based on a deep learning model, comprising: A clause classification identification unit, used to identify the contract clause classification corresponding to each contract clause in the target electronic contract to be managed; the contract clause classification includes at least one of the following: payment clause, breach of contract clause, confidentiality clause, term clause, dispute resolution clause, contract termination clause and liability clause; A key information identification unit, configured to perform risk identification for each of the contract clauses using a clause risk information identification model that matches the contract clause classification of the contract clause, so as to extract key risk information in the contract clause; A clause risk scoring unit, configured to input the key risk information and the contract clause classification matching the contract clause into a risk scoring model to determine a clause risk score corresponding to the contract clause; An interactive control generating unit, configured to generate an audit interactive control according to the contract clause and the key risk information in the contract clause when it is determined that the risk score of the clause exceeds a preset score threshold; the clause risk information identification model and the risk scoring model adopt a deep learning model; An electronic contract archiving unit, configured to determine an archiving tag corresponding to the target electronic contract when all the review interaction controls in the target electronic contract detect a confirmation interaction operation, and classify and store the target electronic contract in an electronic contract archive database according to the archiving tag; The risk scoring model includes an input layer A, a bidirectional LSTM layer, a multi-scale convolutional neural network layer and an output layer B; The input layer A is used to determine the fusion input features corresponding to the key risk information and contract clause classification: ， In the formula, Indicates key risk information m The embedding representation of Embedding representation representing the corresponding contract clause classification, Represents the corresponding fusion input features, and concat represents the feature concatenation operation; The bidirectional LSTM layer is used to capture the contextual information of key risk information and contract terms and generate forward and backward hidden states: ， ， In the formula, and Represents key risk information m The corresponding forward LSTM hidden state and backward LSTM hidden state; The multi-scale convolutional neural network layer uses convolution kernels of different sizes for feature extraction to capture multi-scale feature information: ， ， In the formula, Corresponding key risk information m In the The output feature representation of the convolution kernel is: Indicates key risk information m The corresponding multi-scale feature information; and Respectively represent The weight matrix and bias vector of the convolution kernel, represents the convolution operation, Represents the RELU activation function; The structure of the output layer B is as follows: ， ， ， In the formula, represents the output feature of the fully connected layer, , and Represent the weight matrices of different layers of the output layer B, , and Respectively represent the bias items of different layers of the output layer B; represents the intermediate representation vector, represents the hyperbolic tangent activation function; represents the softmax activation function, Indicates the predicted key risk information m The corresponding clause risk score.

Images