CNN-based large model contract review method in industry cloud environment
By using CNN model to analyze contract text in the industry cloud environment, traditional manual reviews are solved, efficient and accurate contract reviews are achieved, and review efficiency and accuracy are improved, cost is reduced. Enhanced risk assessment capabilities.
Patent Information
- Application Number
- CN202411238788.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-05
- Publication Date
- 2024-12-24
AI Technical Summary
Traditional contract review methods rely on manual review, which is time-consuming and prone to omissions and errors, making it difficult to meet the requirements of modern business activities for the speed and quality of contract review, especially when dealing with long texts and large-scale data.
Using the CNN-based contract review method in the industry cloud environment, the contract text is efficient and accurate analyzed through convolutional neural network, combined with big data and NLP technology, the key terms are automatically identified and their rationality is evaluated, and timely feedback is provided. and suggestions.
It improves the efficiency and accuracy of contract review, reduces manual errors, reduces labor costs, enhances risk assessment capabilities, ensures standardization and consistency, improves user experience, and supports large-scale contract processing.
Smart Images
Figure CN119180275A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence deep learning, and in particular to a large-model contract review method based on CNN in an industry cloud environment. Background Art
[0002] With the development of the global economy and the increasing frequency of commercial activities, contracts have become an indispensable document in various transactions and cooperation. Contract review is an important link to ensure that the content of the contract is reasonable and legal, and is of great significance for avoiding contract disputes and protecting the rights and interests of all parties. However, the traditional contract review method mainly relies on manual review, which takes a lot of time and effort and is prone to omissions and errors.
[0003] In recent years, with the rapid development of artificial intelligence and deep learning technology, it has become possible to use computers to analyze and review contract texts. In particular, convolutional neural networks (CNNs) have achieved remarkable results in areas such as image recognition and natural language processing, providing new ideas and methods for contract review.
[0004] CNN is a special neural network structure with strong feature extraction and classification capabilities. Through the combination of convolutional layers and pooling layers, CNN can automatically learn local and global features in text and convert text into high-dimensional vector representation. This method can not only capture key information in the text, but also identify complex patterns and associations, providing important technical support for contract review.
[0005] However, traditional CNN models have certain limitations when processing long texts and large-scale data. Traditional manual contract review methods often take a lot of time and are prone to errors and omissions due to human subjective judgment and limited energy. In addition, manual review methods are labor-intensive and inefficient when processing a large number of contracts, and it is difficult to meet the requirements of modern business activities for contract review speed and quality. Summary of the invention
[0006] In order to solve the above technical problems, the present invention provides a large-model contract review method based on CNN in an industry cloud environment. By using the powerful neural network structure of convolutional neural network (CNN), the method can achieve efficient and accurate analysis of the contract text and extract the key features in the contract text. It can not only capture the key information in the text, but also identify potential risks and unreasonable terms.
[0007] This method combines big data drive and NLP technology, and uses deep learning technology to efficiently and accurately analyze and review contract texts. It automatically identifies key terms in the contract and evaluates the rationality of the terms, providing users with timely feedback and suggestions.
[0008] The technical solution of the present invention is:
[0009] A large model contract review method based on CNN in an industry cloud environment, including
[0010] Data collection and preprocessing: Collect a large amount of contract text data, clean and preprocess the data, remove irrelevant information, and perform word segmentation and part-of-speech tagging.
[0011] Feature extraction: The CNN model is used to extract features from the contract text. Through the combination of convolutional layers and pooling layers, the local and global features in the contract text are automatically learned, and the text is converted into a high-dimensional vector representation.
[0012] Model training: Use the labeled data set to train the CNN model. Through the back propagation algorithm and gradient descent method, the weights and parameters of the model are optimized so that the model can accurately distinguish between reasonable and unreasonable contract terms.
[0013] Contract review: After the user uploads the contract document, the system uses the trained CNN model to extract and match the contract text. According to the output of the model, the reasonable and unreasonable content in the contract is identified and risk assessment is performed. Output sequence decoding: The predicted word sequence generated by the decoding layer is converted into readable natural language text.
[0014] Results display and suggestions: The review results are displayed to users in a visual way, and unreasonable clauses are highlighted. At the same time, modification suggestions are provided based on the review results to help users improve the contract.
[0015] Furthermore,
[0016] Data collection and preprocessing, including:
[0017] a) Data collection:
[0018] Collect contract text data from various sources, including standard contract templates, historical contracts, and legal provisions;
[0019] b) Data cleaning:
[0020] Remove irrelevant information from the contract text, including headers, footers, signatures, and dates; identify and delete blanks, spaces, and line breaks in the text to purify the data;
[0021] c) Data preprocessing:
[0022] Segment the contract text and divide the continuous text into words or phrases for subsequent feature extraction and model training;
[0023] Perform part-of-speech tagging on the segmented text, marking each word or phrase as a noun, verb, or adjective to provide more contextual information about the words.
[0024] d) Data labeling:
[0025] Annotate the key clauses in the contract text to indicate reasonable and unreasonable clauses; assign a label to each clause for subsequent model training and evaluation.
[0026] Feature extraction specifically includes:
[0027] a) Input preparation:
[0028] Convert the pre-processed contract text data into a format suitable for input into the CNN model;
[0029] b) Word Embedding:
[0030] Use the pre-trained word vector model to map each word or phrase in the contract text into a dense vector representation;
[0031] c) Construct convolutional layer:
[0032] In the CNN model, several convolution kernels are used to perform convolution operations on the input word embedding vector; each convolution kernel captures different types of local features; the convolution operation is performed on the text sequence in a sliding window manner to capture local context information;
[0033] d) Apply activation function:
[0034] After the convolutional layer, a nonlinear activation function is applied to introduce nonlinear characteristics and enhance the expressiveness of the model;
[0035] e) Pooling operation:
[0036] After the convolution layer, a pooling layer is used to reduce the dimensionality of the convolutional features; the pooling operation selects the most important information in the convolutional features and reduces the amount of data to be processed later;
[0037] f) Feature combination:
[0038] Combine the outputs of several convolutional and pooling layers to build a higher-level feature representation;
[0039] g) High-dimensional vector representation:
[0040] Through the combination of several convolution and pooling layers, a high-dimensional vector representation is finally obtained, which contains the local and global features of the contract text.
[0041] in
[0042] Feature combination is achieved by stacking multiple convolutional and pooling layers, with each layer capturing more complex patterns and relationships.
[0043] The high-dimensional vector representation can be used in subsequent classification or regression tasks to evaluate the rationality of the contract terms.
[0044] Model training specifically includes:
[0045] a) Initialize the model:
[0046] Initialize the weights and parameters of the CNN model;
[0047] b) Forward propagation:
[0048] The prepared contract text data is input into the CNN model and processed sequentially through the convolution layer, activation function, and pooling layer. During the forward propagation process, the model will automatically extract features from the contract text and convert them into high-dimensional vector representations.
[0049] c) Calculate the loss:
[0050] Use the loss function to calculate the difference between the model output and the true label; the loss function measures the error between the model's prediction and the actual result, and is used to evaluate the model's performance;
[0051] d) Backward Propagation:
[0052] Based on the calculated loss, the gradients of each weight and parameter in the model are calculated through the back-propagation algorithm; the back-propagation algorithm propagates the error signal from the output layer back to the input layer and calculates the contribution of each weight and parameter to the loss function;
[0053] e) Gradient Descent:
[0054] Update the model's weights and parameters using gradient descent or other optimization algorithms;
[0055] f) Iterative training:
[0056] Repeat the process of forward propagation, loss calculation, back propagation and gradient descent to train the model multiple times. In each iteration, the model will gradually learn the features and patterns in the contract text and improve its ability to distinguish between reasonable and unreasonable contract terms.
[0057] g) Model evaluation:
[0058] During the training process, the validation set or test set is used to evaluate the model regularly. The performance and generalization ability of the model are evaluated by calculating indicators such as the loss and accuracy of the model on the validation set or test set.
[0059] Among them, the gradient descent method adjusts the weights and parameters at the learning rate according to the calculated gradient, so that the loss function value of the model is reduced.
[0060] Contract review, including:
[0061] a) User uploads contract documents:
[0062] Users upload contract documents through the interface or API provided by the system; contract documents can be in various formats, and the system should be able to accept and process documents in different formats;
[0063] b) Document preprocessing:
[0064] Pre-process the contract documents uploaded by users and convert the contract text into a format that the system can process. At the same time, perform preliminary cleaning and word segmentation to remove irrelevant information.
[0065] c) Feature extraction:
[0066] Use the trained CNN model to extract features from the preprocessed contract text; through the combination of convolutional layers and pooling layers, automatically learn the local and global features in the contract text and convert the text into a high-dimensional vector representation;
[0067] d) Feature matching:
[0068] The extracted contract features are matched with the trained model. The model processes the input feature vector through the hidden layer and the fully connected layer, and finally outputs a prediction result indicating the rationality of the contract terms.
[0069] e) Identify reasonable and unreasonable content:
[0070] Based on the output of the model, the reasonable and unreasonable contents in the contract are identified; the model outputs a probability or score, indicating the confidence level that the contract terms are reasonable or unreasonable; based on the preset threshold, the terms are classified as reasonable or unreasonable;
[0071] f) Risk assessment:
[0072] Conduct risk assessment on the unreasonable content identified; evaluate the problems that may be caused by unreasonable terms and provide a risk assessment report.
[0073] The beneficial effects of the present invention are
[0074] Improve efficiency and accuracy: By using a large model based on a convolutional neural network (CNN) for contract review, the efficiency and accuracy of the review can be greatly improved. Automated processing reduces the time required for manual review and reduces the possibility of human error.
[0075] Reduce labor costs: Traditional manual contract review requires a large number of legal professionals to perform time-consuming work, while the method of the present invention can reduce reliance on manual labor, thereby reducing labor costs.
[0076] Enhanced risk assessment capabilities: The method of the present invention can identify and assess the risks of unreasonable content in a contract, helping users to fully understand the potential legal risks and economic impacts of the contract and make more informed decisions.
[0077] Promote standardization and consistency: The automated review process ensures the standardization and consistency of contract reviews, reducing the problem of inconsistent review standards due to differences in personal subjective judgment.
[0078] Improve user experience: Users can obtain quick review results by uploading contract documents, which improves user experience and increases user satisfaction with contract review services.
[0079] Support for large-scale processing: The method of the present invention can process a large amount of contract text data, is suitable for large enterprises or legal service agencies, and meets the needs of large-scale contract review.
[0080] Continuous learning and optimization: Over time, the system of the present invention can continuously optimize the model and improve the accuracy and adaptability of the review by continuously learning from user feedback and market changes.
[0081] Legal compliance: The method of the present invention can help enterprises and individuals better comply with laws and regulations and reduce the occurrence of illegal activities by identifying and modifying non-compliant contract terms. BRIEF DESCRIPTION OF THE DRAWINGS
[0082] Figure 1 It is a schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION
[0083] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0084] A large-model contract review method based on CNN in an industry cloud environment.
[0085] 1) Data collection and preprocessing:
[0086] a) Data collection:
[0087] Collect contract text data from various sources, including but not limited to standard contract templates, historical contracts, legal provisions, etc. Ensure that the collected data is diverse and representative, covering contracts of different industries, fields, and types to enhance the generalization ability of the model.
[0088] b) Data cleaning:
[0089] Remove irrelevant information from the contract text, such as headers, footers, signatures, dates, etc., so that you can focus on the substantive part of the contract content. Identify and delete irrelevant characters such as blanks, spaces, line breaks, etc. in the text to purify the data.
[0090] c) Data preprocessing:
[0091] Segment the contract text and divide the continuous text into words or phrases for subsequent feature extraction and model training.
[0092] Perform part-of-speech tagging on the segmented text, marking each word or phrase as a noun, verb, adjective, etc., to provide more contextual information about the words.
[0093] d) Data labeling:
[0094] Annotate the key clauses in the contract text and mark the reasonable and unreasonable clauses. Assign a label to each clause for subsequent model training and evaluation.
[0095] 2) Feature extraction:
[0096] a) Input preparation:
[0097] Convert the preprocessed contract text data into a format suitable for input to the CNN model. Typically, this involves converting the text data into a fixed-length vector representation, such as by padding or truncation.
[0098] b) Word Embedding:
[0099] Use pre-trained word vector models (such as Word2Vec, GloVe) to map each word or phrase in the contract text into a dense vector representation. This can preserve the semantic information of the vocabulary and reduce the dimensionality curse.
[0100] c) Construct convolutional layer:
[0101] In the CNN model, multiple convolution kernels (filters) are used to convolve the input word embedding vector. Each convolution kernel captures different types of local features, such as keywords, phrases, or patterns. The convolution operation is performed on the text sequence in a sliding window manner to capture local context information.
[0102] d) Apply activation function:
[0103] After the convolutional layer, a nonlinear activation function (such as ReLU) is applied to introduce nonlinear characteristics and enhance the expressiveness of the model.
[0104] e) Pooling operation:
[0105] After the convolutional layer, a pooling layer (such as max pooling or average pooling) is used to reduce the dimensionality of the convolutional features. The pooling operation selects the most important information in the convolutional features and reduces the amount of data for subsequent processing.
[0106] f) Feature combination:
[0107] The outputs of multiple convolutional and pooling layers are combined to build higher-level feature representations. This can be achieved by stacking multiple convolutional and pooling layers, with each layer capturing more complex patterns and relationships.
[0108] g) High-dimensional vector representation:
[0109] Through the combination of multiple convolution and pooling layers, a high-dimensional vector representation is finally obtained, which contains the local and global features of the contract text. This vector representation can be used for subsequent classification or regression tasks to evaluate the rationality of the contract terms.
[0110] 3) Model training:
[0111] a) Initialize the model:
[0112] Initialize the weights and parameters of the CNN model. Usually, the weights can be initialized randomly or using a pre-trained model. Parameters include convolution kernel size, step size, padding method, etc., which are adjusted according to the specific problem and data characteristics.
[0113] b) Forward propagation:
[0114] The prepared contract text data is input into the CNN model and processed sequentially through the convolution layer, activation function, pooling layer, etc. During the forward propagation process, the model will automatically extract the features in the contract text and convert it into a high-dimensional vector representation.
[0115] c) Calculate the loss:
[0116] Use a loss function (such as cross entropy loss) to calculate the difference between the model output and the true label. The loss function measures the error between the model's predicted results and the actual results, and is used to evaluate the performance of the model.
[0117] d) Backward Propagation:
[0118] Based on the calculated loss, the gradients of each weight and parameter in the model are calculated through the back-propagation algorithm. The back-propagation algorithm propagates the error signal from the output layer back to the input layer and calculates the contribution of each weight and parameter to the loss function.
[0119] e) Gradient Descent:
[0120] Use the gradient descent method or other optimization algorithms (such as Adam, RMSprop) to update the weights and parameters of the model. The gradient descent method will adjust the weights and parameters at a certain learning rate based on the calculated gradient, so that the loss function value of the model is reduced, thereby improving the performance of the model.
[0121] f) Iterative training:
[0122] Repeat the process of forward propagation, loss calculation, back propagation and gradient descent to train the model multiple times. In each iteration, the model will gradually learn the features and patterns in the contract text and improve its ability to distinguish between reasonable and unreasonable contract terms.
[0123] g) Model evaluation:
[0124] During the training process, the model is evaluated regularly using the validation set or test set. The performance and generalization ability of the model are evaluated by calculating indicators such as the loss and accuracy of the model on the validation set or test set.
[0125] 4) Contract Review:
[0126] a) User uploads contract documents:
[0127] Users upload contract documents through the interface or API provided by the system. Contract documents can be in various formats, such as PDF, Word, etc. The system should be able to accept and process documents in different formats.
[0128] b) Document preprocessing:
[0129] Pre-process the contract documents uploaded by users, including format conversion, OCR recognition, etc., to convert the contract text into a format that the system can process. At the same time, perform preliminary cleaning and word segmentation to remove irrelevant information, such as headers, footers, signatures, etc.
[0130] c) Feature extraction:
[0131] The trained CNN model is used to extract features from the preprocessed contract text. Through the combination of convolutional layers and pooling layers, the local and global features in the contract text are automatically learned, and the text is converted into a high-dimensional vector representation.
[0132] d) Feature matching:
[0133] The extracted contract features are matched with the trained model. The model processes the input feature vector through a series of hidden layers and fully connected layers, and finally outputs a prediction result indicating the rationality of the contract terms.
[0134] e) Identify reasonable and unreasonable content:
[0135] Based on the output of the model, the reasonable and unreasonable contents in the contract are identified. Usually, the model will output a probability or score, indicating the confidence level that the contract terms are reasonable or unreasonable. Based on the preset threshold, the terms are classified as reasonable or unreasonable.
[0136] f) Risk assessment:
[0137] Conduct risk assessment on the unreasonable contents identified. Based on existing knowledge and experience, evaluate the legal consequences and economic impacts that may result from unreasonable clauses. Provide users with detailed risk assessment reports to help them fully understand the potential risks in the contract.
[0138] 5) Results and suggestions
[0139] a) The review results are displayed to the user in a visual way, highlighting unreasonable clauses. At the same time, modification suggestions are provided based on the review results to help users improve the contract.
[0140] The present invention aims to provide an auxiliary tool to help users discover potential risks and unreasonable terms before signing a contract. By automatically identifying key terms in a contract and evaluating the rationality of the terms, the present invention aims to provide users with timely feedback and suggestions, reduce the risk of legal disputes, and protect the legitimate rights and interests of all parties.
[0141] Ultimately, the purpose of this invention is to improve the overall efficiency and effectiveness of contract management by improving the automation level of contract review. By introducing deep learning technology and big data driven methods, this invention simplifies the contract review process and speeds it up, providing users with an efficient and accurate contract review tool. This will help users better manage contracts, reduce contract risks, improve transaction efficiency, and further promote the prosperity and development of commercial activities.
[0142] The present invention uses deep learning technology, especially the CNN model, to efficiently analyze and review the contract text, which can improve the accuracy and efficiency of contract review. At the same time, this method can be applied to different types of contracts and has good generalization ability. However, due to the complexity of legal provisions and contract terms, the AI review results can be used as a reference for professional legal knowledge when conducting reviews.
[0143] The above description is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.
Claims
1. A large model contract review method based on CNN in an industry cloud environment, characterized in that: include: (1) Data collection and preprocessing: Collect contract text data, clean and preprocess the data, and perform word segmentation and part-of-speech tagging; (2) Feature extraction: The CNN model is used to extract features from the contract text. Through the combination of convolutional layers and pooling layers, the local and global features in the contract text are automatically learned and the text is converted into a high-dimensional vector representation. (3) Model training: Use the labeled data set to train the CNN model; optimize the model weights and parameters through the back-propagation algorithm and gradient descent method so that the model can distinguish between reasonable and unreasonable contract terms; (4) Contract review: After the user uploads the contract document, the trained CNN model is used to extract and match the contract text. Based on the output of the model, the reasonable and unreasonable contents in the contract are identified and risk assessment is performed. (5) Result display and suggestions: The review results are displayed to the user in a visual manner. At the same time, modification suggestions are provided based on the review results to help users improve the contract.
2. The method according to claim 1, characterized in that Data collection and preprocessing, including: a) Data collection: Collect contract text data from various sources, including standard contract templates, historical contracts, and legal provisions; b) Data cleaning: Remove irrelevant information from the contract text, including headers, footers, signatures, and dates; identify and delete blanks, spaces, and line breaks in the text to purify the data; c) Data preprocessing: Segment the contract text and divide the continuous text into words or phrases for subsequent feature extraction and model training; Perform part-of-speech tagging on the text after tokenization, marking each word or phrase as a noun, verb, or adjective to provide more contextual information about the words; d) Data labeling: Annotate the key clauses in the contract text to indicate reasonable and unreasonable clauses; assign a label to each clause for subsequent model training and evaluation.
3. The method according to claim 1, characterized in that Feature extraction specifically includes: a) Input preparation: Convert the pre-processed contract text data into a format suitable for input into the CNN model; b) Word Embedding: Use the pre-trained word vector model to map each word or phrase in the contract text into a dense vector representation; c) Construct convolutional layer: In the CNN model, several convolution kernels are used to perform convolution operations on the input word embedding vector; each convolution kernel captures different types of local features; the convolution operation is performed on the text sequence in a sliding window manner to capture local context information; d) Apply activation function: After the convolutional layer, a nonlinear activation function is applied to introduce nonlinear characteristics and enhance the expressiveness of the model; e) Pooling operation: After the convolution layer, a pooling layer is used to reduce the dimensionality of the convolutional features; the pooling operation selects the most important information in the convolutional features and reduces the amount of data to be processed later; f) Feature combination: Combine the outputs of several convolutional and pooling layers to build a higher-level feature representation; g) High-dimensional vector representation: Through the combination of several convolution and pooling layers, a high-dimensional vector representation is finally obtained, which contains the local and global features of the contract text.
4. The method according to claim 3, characterized in that Feature combination is achieved by stacking multiple convolutional and pooling layers, with each layer capturing more complex patterns and relationships.
5. The method according to claim 3, characterized in that: The high-dimensional vector representation can be used in subsequent classification or regression tasks to evaluate the rationality of the contract terms.
6. The method according to claim 3, characterized in that Model training specifically includes: a) Initialize the model: Initialize the weights and parameters of the CNN model; b) Forward propagation: The prepared contract text data is input into the CNN model and processed sequentially through the convolution layer, activation function, and pooling layer. During the forward propagation process, the model will automatically extract features from the contract text and convert them into high-dimensional vector representations. c) Calculate the loss: Use the loss function to calculate the difference between the model output and the true label; the loss function measures the error between the model's prediction and the actual result, and is used to evaluate the model's performance; d) Backward Propagation: Based on the calculated loss, the gradients of each weight and parameter in the model are calculated through the back-propagation algorithm; the back-propagation algorithm propagates the error signal from the output layer back to the input layer and calculates the contribution of each weight and parameter to the loss function; e) Gradient Descent: Update the model's weights and parameters using gradient descent or other optimization algorithms; f) Iterative training: Repeat the process of forward propagation, loss calculation, back propagation and gradient descent to train the model multiple times. In each iteration, the model will gradually learn the features and patterns in the contract text and improve its ability to distinguish between reasonable and unreasonable contract terms. g) Model evaluation: During the training process, the validation set or test set is used to evaluate the model regularly. The performance and generalization ability of the model are evaluated by calculating indicators such as the loss and accuracy of the model on the validation set or test set.
7. The method according to claim 6, characterized in that The gradient descent method adjusts weights and parameters at a learning rate based on the calculated gradient, so that the loss function value of the model decreases.
8. The method according to claim 1, characterized in that: Contract review, including: a) User uploads contract documents: Users upload contract documents through the interface or API provided by the system; contract documents can be in various formats, and the system should be able to accept and process documents in different formats; b) Document preprocessing: Pre-process the contract documents uploaded by users and convert the contract text into a format that the system can process. At the same time, perform preliminary cleaning and word segmentation to remove irrelevant information. c) Feature extraction: Use the trained CNN model to extract features from the preprocessed contract text; through the combination of convolutional layers and pooling layers, automatically learn the local and global features in the contract text and convert the text into a high-dimensional vector representation; d) Feature matching: The extracted contract features are matched with the trained model. The model processes the input feature vector through the hidden layer and the fully connected layer, and finally outputs a prediction result indicating the rationality of the contract terms. e) Identify reasonable and unreasonable content: Based on the output of the model, the reasonable and unreasonable contents in the contract are identified; the model outputs a probability or score, indicating the confidence level that the contract terms are reasonable or unreasonable; based on the preset threshold, the terms are classified as reasonable or unreasonable; f) Risk assessment: Conduct risk assessment on the unreasonable content identified; evaluate the problems that may be caused by unreasonable terms and provide a risk assessment report.