Contract text structured processing method and device, equipment and storage medium

Abstract

The invention provides a structured processing method and device for a contract text, equipment and a storage medium, and the method comprises the steps: carrying out the preprocessing of the contract text, generating a standardized word sequence, and enabling the standardized word sequence to comprise a plurality of word sequences and part-of-speech tagging information of each word sequence; the standardized word sequence is input into a named entity recognition model to recognize key entities in the contract text, key terms in the contract text are extracted based on a machine learning algorithm, the key entities include both parties of the contract, the amount of the contract and the name of a product, and the key terms include payment terms, delivery terms, default responsibility terms and force majeure terms; through dependency syntactic analysis and a semantic role labeling model, analyzing to obtain a semantic relationship between the key entities and the key terms; and storing the extracted key entities, key terms and semantic relationships in a target database in a structured form, and establishing a target index to support quick query. According to the invention, the error rate of manual processing is reduced.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a structured processing method, device, equipment and storage medium for contract text. Background Art

[0002] Contracts are crucial legal documents that protect the rights and interests of all parties. With the acceleration of digitalization, a large number of contracts exist in electronic form. However, contracts are typically unstructured or semi-structured natural language texts, containing a wealth of information such as terms, conditions, responsibilities, and obligations, making them difficult for computer systems to directly process and analyze.

[0003] At present, the structured processing of contract texts mainly relies on manual information extraction and organization. This method is not only inefficient and error-prone, but also difficult to quickly process large-scale contract texts.

[0004] Therefore, an efficient and accurate contract text structured processing method is needed to improve the efficiency of contract management and analysis. Summary of the Invention

[0005] The present invention provides a method, device, equipment and storage medium for structured processing of contract texts, which are used to solve the defects of low efficiency and easy error in structured processing in the existing technology, reduce the error rate of manual processing, and thus improve the efficiency of contract management and analysis.

[0006] In a first aspect, the present invention provides a method for structured processing of a contract text, the method comprising the following steps: Preprocessing the contract text to generate a standardized word sequence; the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; Inputting the standardized word sequence into a named entity recognition model to identify key entities in the contract text, and extracting key terms in the contract text based on a machine learning algorithm; the key entities include the contracting parties, the contract amount, and the product name; and the key terms include payment terms, delivery terms, breach of contract liability terms, and force majeure terms; Analyzing the key entity and the key clause by using dependency syntax, and inputting the analysis results into a semantic role labeling model to obtain a semantic relationship between the key entity and the key clause; the dependency syntax is used to identify the grammatical structure between word sequences; The extracted key entities, key terms, and semantic relationships are stored in a structured form in a target database, and a target index is established to support fast query.

[0007] According to a structured processing method for a contract text provided by the present invention, the contract text is preprocessed to generate a standardized word sequence, including: removing special characters, redundant spaces, and line breaks from the contract text to obtain a cleaned contract text; Performing word segmentation processing on the cleaned contract text using an optimized word segmentation algorithm based on a contract domain dictionary to obtain multiple word sequences; Part-of-speech tagging is performed on each of the word sequences to obtain part-of-speech tagging information of each word sequence; the part-of-speech tagging information includes nouns, verbs, and adjectives.

[0008] According to a method for structured processing of a contract text provided by the present invention, inputting the standardized word sequence into a named entity recognition model to identify key entities in the contract text includes: Performing character-level encoding on each of the word sequences to generate an input tensor corresponding to each of the word sequences including a word vector, a positional encoding, and a segment encoding; Inputting each of the input tensors into the named entity recognition model to extract each contextual semantic feature; Outputting entity label probability distribution information for each of the word sequences based on each of the contextual semantic features, and determining a target label sequence corresponding to each of the word sequences based on the entity label probability distribution information for each of the word sequences; the target label sequence is used to characterize the entity type corresponding to the word sequence; Continuous entity segments are merged according to target tag sequences corresponding to the word sequences to generate a structured entity list, and the structured entity list is determined as the identified key entities.

[0009] According to a method for structured processing of a contract text provided by the present invention, after inputting the standardized word sequence into a named entity recognition model to identify key entities in the contract text, the method further includes: Verify the contract domain rules on the identified structured entity list; If there are multiple candidate results for the same entity, ambiguity resolution is performed based on context confidence and rule priority.

[0010] According to a structured processing method for a contract text provided by the present invention, extracting key terms from the contract text based on a machine learning algorithm includes: Based on a preset keyword library of terms, a bidirectional long short-term memory network model is used to classify and extract the terms to obtain the payment terms, the delivery terms, the breach of contract liability terms and the force majeure terms; The extracted key terms are subjected to confidence check. If the confidence is lower than the threshold, secondary matching and correction are performed in combination with the rule template.

[0011] According to a structured processing method for a contract text provided by the present invention, the method includes analyzing the key entities and the key clauses by dependency syntax, inputting the analysis results into a semantic role labeling model, and obtaining the semantic relationship between the key entities and the key clauses, including: Performing dependency syntax analysis on the sentence containing the key entity to generate a syntax tree to identify the modification relationship between each of the word sequences; Inputting the syntax tree into the semantic role labeling model to obtain the semantic role of each word sequence; the semantic role of each word sequence is used to determine the subject, action, object relationship, condition and result relationship; According to the semantic roles of each of the word sequences, the contract subject layer, the obligation clause layer, the breach of contract clause layer and the force majeure clause layer are connected through directed edges to form a hierarchical association network.

[0012] According to a structured processing method for a contract text provided by the present invention, the extracted key entities, key terms, and semantic relationships are stored in a structured form in a target database, and a target index is established to support fast query, including: If the target database is a relational database, the key entity is mapped to an entity table field in the relational database, and the semantic relationship is mapped to a foreign key constraint of an association table; If the target database is a graph database, the key entities are used as nodes and the semantic relationships are used as edges to construct graph structure data; The target index is generated based on key entity names, clause types and semantic relationship types; the target index is a multi-dimensional joint index, and the target index supports compound queries based on entity attributes, clause content and relationship paths.

[0013] In a second aspect, the present invention further provides a structured processing device for a contract document, the device comprising the following modules: A preprocessing module is used to preprocess the contract text to generate a standardized word sequence; the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; an information extraction module, configured to input the standardized word sequence into a named entity recognition model, identify key entities in the contract text, and extract key terms in the contract text based on a machine learning algorithm; the key entities include the contracting parties, the contract amount, and the product name; and the key terms include payment terms, delivery terms, breach of contract liability terms, and force majeure terms; A semantic analysis module, configured to analyze the key entities and the key terms using dependency syntax, and input the analysis results into a semantic role labeling model to obtain the semantic relationship between the key entities and the key terms; the dependency syntax is used to identify the grammatical structure between word sequences; The storage module stores the extracted key entities, key terms and semantic relationships in a structured form in a target database, and establishes a target index to support fast query.

[0014] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the processor implements a structured processing method for a contract text as described above.

[0015] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the structured processing method of the contract text as described in any one of the above.

[0016] In a fifth aspect, the present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described structured processing methods for contract texts.

[0017] The present invention provides a structured processing method, device, equipment and storage medium for contract texts. First, the contract text is preprocessed to generate a standardized word sequence, wherein the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; then, the standardized word sequence is input into a named entity recognition model to identify key entities in the contract text, and key terms in the contract text are extracted based on a machine learning algorithm. The key entities include the two parties to the contract, the contract amount and the product name, and the key terms include payment terms, delivery terms, breach of contract liability terms and force majeure terms. Furthermore, through dependency syntax analysis and semantic role labeling models, the semantic relationships between key entities and key terms are parsed; then, the extracted key entities, key terms and semantic relationships are stored in a target database in a structured form, and a target index is established to support fast query.

[0018] The present invention uses automated text processing and information extraction technology, and utilizes pre-trained models and machine learning algorithms to extract key information, obtain extracted key entities and key clauses, and use semantic role labeling models to parse the semantic relationships between key entities and key clauses. It can accurately identify key information and semantic relationships in the contract, and store the key information and semantic relationships in a structured manner, reducing the error rate of manual processing and thereby improving the efficiency of contract management and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is 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 invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 This is one of the flow charts of the structured processing method for contract text provided by the present invention.

[0021] Figure 2 This is the second flow chart of the structured processing method of the contract text provided by the present invention.

[0022] Figure 3 It is a schematic diagram of the structure of the structured processing device for contract text provided by the present invention.

[0023] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0024] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0025] In order to more clearly understand the various embodiments provided by the present invention, the technical content of the present invention is first introduced as follows: There is a need for an efficient and accurate contract text structuring method to improve the efficiency of contract management and analysis.

[0026] Some existing automated processing methods have limited ability to understand and structure complex contract texts, and are unable to accurately identify key information and semantic relationships within contracts. Traditional regular expression matching cannot handle the complex semantic relationships within contract clauses (such as the nesting of "force majeure" clauses and multi-paragraph liability exemption clauses). Existing solutions often rely on keyword extraction and lack the ability to reconstruct hierarchical structures. The accuracy of parsing non-text elements such as tables and seals in Portable Document Format (PDF) and scanned documents is insufficient.

[0027] Based on the deficiencies in the existing technology, the present invention provides a method, device, equipment and storage medium for structured processing of contract texts, which can efficiently and accurately structure contract texts, extract key information in the contract, establish semantic relationships between information, and thereby improve the efficiency of contract management and analysis.

[0028] The following combination Figure 1-Figure 4 The present invention describes a method, apparatus, device and storage medium for structured processing of contract texts.

[0029] Figure 1 This is one of the flow charts of the structured processing method for the contract text provided by the present invention, such as Figure 1 As shown, the method includes the following: Step 101: Preprocess the contract text to generate a standardized word sequence; the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; Specifically, it should be noted that the execution subject of the present invention is an electronic device, which is used to implement structured processing of the contract text.

[0030] First, the contract text is obtained. In this embodiment, contracts include multiple categories, such as evaluation contracts, assessment contracts, certification contracts, etc.

[0031] Furthermore, before formally structuring the contract text, it undergoes preprocessing. This includes cleaning noise, segmenting words, and tagging parts of speech. The goal of preprocessing is to convert the contract text into a sequence of words that is easy to process and to clarify the parts of speech of the words. This means generating standardized word sequences, which include multiple word sequences and part-of-speech tagging information for each. This lays a solid foundation for subsequent key information extraction and semantic relationship analysis, ensuring the model focuses on meaningful text content.

[0032] Step 102: Input the standardized word sequence into a named entity recognition model to identify key entities in the contract text, and extract key terms in the contract text based on a machine learning algorithm; key entities include the contracting parties, contract amount, and product name; key terms include payment terms, delivery terms, breach of contract liability terms, and force majeure terms; Specifically, after preprocessing to obtain standardized word sequences (i.e., multiple word sequences and part-of-speech tagging information for each word sequence), the system further combines pre-trained Named Entity Recognition (NER) models, rule matching, and machine learning algorithms to extract key information from contract texts.

[0033] First, the standardized word sequence is input into the named entity recognition model to identify key entities in the contract text. Key entities include the contracting parties, contract date, contract amount, product name, etc. The named entity recognition model processing process typically includes label encoding, feature extraction, model training, model recognition, and post-processing.

[0034] Furthermore, key terms from the contract text are extracted based on a machine learning algorithm. These key terms include payment terms, delivery terms, breach of contract liability clauses, and force majeure clauses. A machine learning algorithm is a mathematical model that allows computers to automatically learn patterns from data and use it for predictions or decision-making.

[0035] The core types of machine learning algorithms include: 1. Supervised learning: Random Forest, Bi-directional Long Short-Term Memory-Conditional Random Field (BiLSTM-CRF); 2. Unsupervised Learning: Clustering (K-means) 3. Deep Learning: Transformers: Attention is All You Need: A new neural network structure based on the self-attention mechanism and graph neural networks.

[0036] The steps to extract key contract terms are as follows: (1) Data collection: corporate legal affairs system, public judgment document network, and manually annotated contracts Format processing: Portable Document Format (PDF) parsing, scanned optical character recognition (OCR) format conversion.

[0037] (2) Data annotation: formulation of annotation standards: Types of clauses: payment terms, delivery terms, breach of contract liability clauses and force majeure clauses (3) Text vectorization: (4) Machine learning algorithms identify key terms.

[0038] Step 103: Analyze key entities and key terms using dependency syntax, input the analysis results into a semantic role labeling model, and obtain semantic relationships between key entities and key terms; dependency syntax is used to identify grammatical structures between word sequences; Specifically, after extracting key information (key entities and key terms), further, dependency syntax can be used to analyze the sentences containing the key entities and key terms to obtain analysis results (such as a grammatical tree of the sentences containing the key entities and key terms).

[0039] Dependency syntax explains the syntactic structure of a language unit by analyzing the dependency relationships between components within the unit. It argues that the core verb in a sentence is the central component that controls other components. However, it itself is not controlled by any other component. All controlled components are subordinate to the controller in some way.

[0040] The grammatical tree is then fed into a semantic role labeling (SRL) model to identify the semantic relationships (i.e., semantic roles) between key entities and key clauses. Dependency parsing reveals the grammatical structure between words, helping to understand the basic framework of the sentence. The SRL model further explores the semantic relationships between key pieces of information, such as subject-action-object relationships and condition-result relationships, to provide a deeper understanding of the meaning of the contract terms.

[0041] Step 104: Store the extracted key entities, key terms, and semantic relationships in a structured form in a target database, and create a target index to support fast query.

[0042] Specifically, the extracted key entities, key terms, and semantic relationships can be stored in a structured form in a target database, and a target index can be established to support fast query.

[0043] The target database is, for example, a graph database or a relational database, and the target index is a multi-dimensional structured joint index. The establishment of the index includes a multi-dimensional joint index based on key entity names, term types, and semantic relationships.

[0044] The method provided in this embodiment first preprocesses the contract text to generate a standardized word sequence, wherein the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; then, the standardized word sequence is input into a named entity recognition model to identify key entities in the contract text, and key terms in the contract text are extracted based on a machine learning algorithm. The key entities include the two parties to the contract, the contract amount and the product name, and the key terms include payment terms, delivery terms, breach of contract liability terms and force majeure terms. Furthermore, through dependency syntax analysis and semantic role labeling models, the semantic relationships between the key entities and key terms are parsed; then, the extracted key entities, key terms and semantic relationships are stored in a target database in a structured form, and a target index is established to support fast query.

[0045] The present invention uses automated text processing and information extraction technology, and utilizes pre-trained models and machine learning algorithms to extract key information, obtain extracted key entities and key clauses, and use semantic role labeling models to parse the semantic relationships between key entities and key clauses. It can accurately identify key information and semantic relationships in the contract, and store the key information and semantic relationships in a structured manner, reducing the error rate of manual processing and thereby improving the efficiency of contract management and analysis.

[0046] It should be noted that each implementation method of the present application can be freely combined, the order can be changed, or it can be executed separately, and does not need to rely on or depend on a fixed execution order.

[0047] According to a structured processing method for a contract text provided by the present invention, the contract text is preprocessed to generate a standardized word sequence, including: Remove special characters, redundant spaces, and line breaks from the contract text to obtain the cleaned contract text; An optimized word segmentation algorithm based on the contract domain dictionary is used to segment the cleaned contract text to obtain multiple word sequences; Part-of-speech tagging is performed on each word sequence to obtain part-of-speech tagging information of each word sequence; the part-of-speech tagging information includes nouns, verbs and adjectives.

[0048] Specifically, in some embodiments, the text preprocessing process in step 101 is implemented by the following steps: First, remove special characters, redundant spaces, and line breaks from the contract text to obtain a cleaned contract text, thus eliminating noise. For example, special characters, redundant spaces, and line breaks can be removed by matching regular expressions. Regular expressions (Regular Expressions, or Regex) are patterns used to match character combinations within a string. They are widely used in string search, replacement, and segmentation operations. In Python, a cross-platform, object-oriented, and dynamically typed computer programming language, we can use the re module to process regular expressions.

[0049] Furthermore, the cleaned contract text is segmented using an optimized word segmentation algorithm based on a contract-related dictionary, resulting in multiple word sequences. The contract-related dictionary is a pre-defined vocabulary library. The multi-source dictionary construction and update process includes the collection of legal terminology, manual verification, sample output of dictionary segments, real-time storage of new user-annotated terms, and maintenance of the dictionary change history, with support for rollbacks.

[0050] After obtaining the cleaned text, we further segment it using an optimized segmentation algorithm based on the contract domain dictionary to obtain multiple word sequences. Examples of segmentation algorithms include Jieba, a Python Chinese word segmentation component based on dictionary matching, a sequence tagging model (Long-term Potentiation, LTP), and a deep learning model (HanLP-Bidirectional Encoder Representations from Transformers, HanLP-BERT).

[0051] For example, the word segmentation process includes basic word segmentation, regular expression correction, legal term verification, etc.

[0052] #Example tokenizer = LegalTokenizer() Text ="Party A shall pay USD 5 million to the designated account in §3.2 before December 31, 2024" tokens =tokenizer.tokenize(text) # Output: ['Party A', 'Should', 'On', 'December 31, 2024', 'Before', 'Payment', 'USD 5 million', 'To', '§3.2', 'Designated account'].

[0053] Furthermore, part-of-speech tagging is performed on each word sequence to obtain part-of-speech tagging information of each word sequence. The part-of-speech tagging information includes nouns, verbs, and adjectives.

[0054] The method provided in this embodiment comprehensively utilizes operations such as cleaning, word segmentation, and part-of-speech tagging to remove noise information from the contract text, convert the text into a word sequence that is convenient for subsequent processing, and clarify the part of speech of the words, thereby laying the foundation for subsequent key information extraction and semantic analysis, ensuring that the model can focus on meaningful text content, improving the accuracy of model recognition, and thus more effectively managing contract texts.

[0055] According to a structured processing method for a contract text provided by the present invention, a standardized word sequence is input into a named entity recognition model to identify key entities in the contract text, including: Perform character-level encoding on each word sequence to generate an input tensor corresponding to each word sequence, including word vectors, positional encoding, and segment encoding; Input each input tensor into the named entity recognition model to extract the contextual semantic features; Output entity label probability distribution information for each word sequence based on the semantic features of each context, and determine the target label sequence corresponding to each word sequence based on the entity label probability distribution information of each word sequence; the target label sequence is used to represent the entity type corresponding to the word sequence; Continuous entity segments are merged according to the target label sequences corresponding to each word sequence to generate a structured entity list, and the structured entity list is determined as the key entity for recognition.

[0056] Specifically, in some embodiments, step 102 of extracting key information can be implemented by the following steps: The present invention adopts a domain-adaptive NER model based on the BERT architecture. The model is pre-trained in two stages on a general corpus (such as Wikipedia and news text) and a contract domain annotated dataset, where the contract domain dataset contains manually annotated entities of the contract parties, dates, amounts, and product names.

[0057] Load the pre-trained model parameters and the contract domain entity label mapping table. The label mapping table defines the correspondence between entity types and contract terms (for example, "Party A" corresponds to "PARTY_A" and "Contract Amount" corresponds to "CONTRACT_AMOUNT").

[0058] Key information extraction is achieved through the following steps: First, character-level encoding is performed on each word sequence to generate an input tensor corresponding to each word sequence, which contains word vectors, positional encoding, and segment encoding to adapt to the input requirements of the named entity recognition model (BERT model).

[0059] Furthermore, each input tensor is fed into a named entity recognition model to extract contextual semantic features. For example, contextual semantic features are extracted using a multi-head self-attention mechanism in a named entity recognition (NER) model.

[0060] Furthermore, the entity label probability distribution information of each word sequence is output based on the semantic features of each context, and based on the entity label probability distribution information of each word sequence, the target label sequence corresponding to each word sequence is determined, and the target label sequence is used to characterize the entity type corresponding to the word sequence.

[0061] That is, the entity label probability distribution of each word sequence is output, and the conditional random field (CRF) decoding layer is used to optimize the label sequence to determine the entity's start (B), internal (I) and non-entity (O) labels.

[0062] Furthermore, continuous entity segments are merged according to the target tag sequences corresponding to each word sequence to generate a structured entity list, and the structured entity list is determined as the key entity to be identified. For example, the structured entity list generated is "Party A: XXX Company".

[0063] The method provided in this embodiment combines a pre-trained named entity recognition (NER) model with rule matching and machine learning algorithms. The NER model can learn rich language features and effectively identify key entities in contracts.

[0064] According to a structured processing method for a contract text provided by the present invention, after inputting a standardized word sequence into a named entity recognition model and identifying key entities in the contract text, the method further includes: Verify the contract domain rules on the identified structured entity list; If there are multiple candidate results for the same entity, ambiguity resolution is performed based on context confidence and rule priority.

[0065] Specifically, in some embodiments, after completing the key information extraction in step 102, the following steps are further included: The identified structured entity list is verified according to the contract domain rules.

[0066] For example, if the entities of both parties to the contract are missing, they are completed by matching the "Party A / Party B" prefix or the text in the signature area of the contract signing page through regular expressions; for another example, the unit of the contract amount entity is unified (such as "RMB One Million Yuan" is converted to "1,000,000 Yuan"); for another example, the product name entity is normalized by synonyms (such as "laptop" and "laptop" are mapped to a unified term).

[0067] If there are multiple candidate results for the same entity, ambiguity resolution is performed based on context confidence and rule priority.

[0068] For example, if there are multiple candidate results for the same entity, ambiguity is resolved based on context confidence and rule priority (for example, the amount entity close to the "signing date" is given priority).

[0069] The method provided in this embodiment, after identifying the key entities, further performs contract domain rule verification and ambiguity resolution on the structured entity list, thereby improving the accuracy of key information extraction.

[0070] According to the present invention, a structured processing method for a contract text is provided, which extracts key clauses from the contract text based on a machine learning algorithm, including: Based on the preset keyword library, a bidirectional long short-term memory network model is used for classification and extraction to obtain payment terms, delivery terms, breach of contract liability terms, and force majeure terms; The extracted key terms are checked for confidence. If the confidence is lower than the threshold, secondary matching and correction are performed in combination with the rule template.

[0071] Specifically, in some embodiments, the specific implementation process of extracting key terms from the contract text based on the machine learning algorithm in step 102 includes the following steps: First, based on the preset keyword library of terms, a bidirectional long short-term memory network model is used for classification and extraction to obtain payment terms, delivery terms, breach of contract liability terms and force majeure clauses.

[0072] Among them, the preset clause keyword library is obtained by summarizing a large amount of contract data, which includes keywords corresponding to various types of key clauses.

[0073] The Bidirectional Long Short-Term Memory (BiLSTM) network is an improved Recurrent Neural Network (RNN) specifically designed for processing sequential data. The BiLSTM consists of two parallel LSTM layers: one processes the forward portion of the input sequence, computing sequentially from the first to the last element; the other processes the reverse portion, computing in reverse order from the last to the first element. The outputs of these two layers are merged at each time step or at the end of the sequence to enhance the model's ability to capture bidirectional context. The BiLSTM uses a forward LSTM and a backward LSTM to obtain two hidden layer states with opposite time series, then concatenates them to produce a single output. The remaining steps are similar to the LSTM training process. The forward LSTM and backward LSTM layers can obtain forward and backward information about the current input sequence, respectively.

[0074] Based on the preset keyword library, the contract text is input into the bidirectional long short-term memory network model to obtain various types of key terms that are classified and extracted. The process is as follows: (1) Keyword annotation: According to the preset keyword library, the keywords in the text are annotated so that the model can recognize the characteristics of different terms.

[0075] (2) Model construction: Embedding layer: Converts text into word vector representation. You can use pre-trained word vectors (such as GloVe) or train your own word vectors.

[0076] Bidirectional LSTM layer: Builds a bidirectional LSTM network to capture contextual information in the text. The forward LSTM processes the forward information of the text, while the backward LSTM processes the backward information of the text. Finally, the outputs of the two are concatenated.

[0077] Fully connected layer: A fully connected layer is added after the bidirectional LSTM layer to map the output of the LSTM to the classification label space.

[0078] Output layer: According to the requirements of the classification task, select a suitable activation function (such as softmax) to output the classification results.

[0079] (3) Model training: The preprocessed data is input into the model and multiple rounds of iterative training are performed until the performance of the model reaches the expected level.

[0080] (4) Model identification: By inputting the contract text into the bidirectional long short-term memory network model, various types of key terms can be classified and extracted, including payment terms, delivery terms, breach of contract liability terms and force majeure clauses.

[0081] Afterwards, the extracted key terms can be further verified for confidence. If the confidence is below the threshold, a secondary match and correction is performed in conjunction with the rule template. Ultimately, the key terms in the contract text can be accurately extracted.

[0082] The method provided in this embodiment uses a bidirectional long short-term memory network model to classify and extract key terms based on a preset keyword library to obtain payment terms, delivery terms, breach of contract liability terms, and force majeure terms. The extracted key terms are then confidence-checked. If the confidence is lower than a threshold, secondary matching and correction are performed in combination with a rule template, thereby improving the accuracy of key term extraction.

[0083] According to a structured processing method for a contract text provided by the present invention, the key entities and the key clauses are analyzed by dependency syntax, and the analysis results are input into a semantic role labeling model to obtain the semantic relationship between the key entities and the key clauses, including: Perform dependency syntax analysis on the sentences containing key entities and generate a grammatical tree to identify the modification relationship between each word sequence; Input the syntax tree into the semantic role labeling model to obtain the semantic role of each word sequence; the semantic role of each word sequence is used to determine the subject, action, object relationship, condition and result relationship; According to the semantic role of each word sequence, the contract body layer, obligation clause layer, breach of contract clause layer and force majeure clause layer are connected through directed edges to form a hierarchical association network.

[0084] Optionally, after identifying key entities and key terms, the identified entities can be associated with the key terms (for example, binding the "contract amount" entity to the "payment terms"); output a structured entity list and association relationships as input for semantic relationship analysis.

[0085] In some embodiments, the specific implementation process of obtaining the semantic relationship between the key entity and the key terms by model recognition in step 103 includes the following steps: First, the sentence containing the key entity is subjected to dependency syntactic analysis, and a grammatical tree is generated to identify the modification relationship between each word sequence.

[0086] Among them, the five conditions for adopting dependency syntax include: (1) Only one component in a sentence is independent; (2) All other elements of the sentence are subordinate to a certain element; (3) No single component can be dependent on two or more components; (4) If component A is directly subordinate to component B, and component C is located between A and B in the sentence, then component C is either subordinate to A, or to B, or to a component between A and B; (5) The other components on the left and right sides of the central component have no relationship with each other.

[0087] The marking relationships include, for example, subject-predicate relationship, verb-object relationship, indirect-object relationship, prepositioned object, attributive-predicate relationship, parallel relationship, preposition-object relationship, left-attachment relationship, right-attachment relationship, etc.

[0088] Furthermore, the grammatical tree is input into the semantic role labeling model to obtain the semantic role of each word sequence. The semantic role of each word sequence is used to determine the subject, action, object relationship, condition and result relationship.

[0089] The semantic role labeling model is, for example, a deep learning-based model, which is obtained by training an initial network model with semantic role labeling information of multiple sample grammar trees and corresponding word sequences.

[0090] By inputting the syntax tree, the semantic roles of each word sequence can be identified, that is, the semantic relationships can be identified.

[0091] Furthermore, based on the semantic roles of each word sequence, the contract subject layer, the obligation clause layer, the breach of contract clause layer, and the force majeure clause layer are connected by directed edges to form a hierarchical association network. The hierarchical association network includes the hierarchical association relationship between the contract subject layer, the obligation clause layer, and the breach of contract clause layer.

[0092] The method provided in this embodiment first performs dependency syntax analysis on the sentence containing the key entity, generates a syntax tree to identify the modification relationship between each word sequence, then inputs the syntax tree into the semantic role annotation model to obtain the semantic role of each word sequence. The semantic role of each word sequence is used to determine the subject, action, object relationship, condition, and result relationship; then, based on the semantic role of each word sequence, the contract subject layer, obligation clause layer, breach of contract clause layer, and force majeure clause layer are connected through directed edges to form a hierarchical association network. The SRL model can annotate the semantic roles in the contract text and clarify the roles and relationships of each entity in the sentence. This helps to understand the specific meaning and constraints of the contract terms. For example, in the payment terms, key information such as the payer, payee, payment amount, and payment time and their mutual relationships can be accurately identified, thereby improving the accuracy and efficiency of contract management.

[0093] According to the present invention, a method for structured processing of contract texts is provided, which stores the extracted key entities, key terms, and semantic relationships in a structured form in a target database and establishes a target index to support fast query, including: If the target database is a relational database, map the key entities to entity table fields in the relational database, and map the semantic relationships to foreign key constraints of the associated tables; If the target database is a graph database, key entities are used as nodes and semantic relationships as edges to construct graph-structured data; Generate a target index based on key entity names, clause types, and semantic relationship types; the target index is a multi-dimensional joint index that supports compound queries based on entity attributes, clause content, and relationship paths.

[0094] Specifically, in some embodiments, the storage query step in step 102 may be implemented in the following manner: If the target database is a relational database, the key entities are mapped to entity table fields in the relational database, and the semantic relationships are mapped to foreign key constraints of the associated tables.

[0095] A relational database is a database management system that uses the relational model to organize and manage data. The relational model consists of three parts: a relational data structure, a set of relational operations, and relational integrity constraints. The relational model divides data into tables, each containing a set of records, each containing a set of attributes. Relationships can be established between these tables, allowing data to be shared across multiple tables.

[0096] In this embodiment, in a relational database for storing key entities, key terms, and semantic relationships, key entities are mapped to entity table fields, and semantic relationships are mapped to foreign key constraints of associated tables.

[0097] If the target database uses a graph database, key entities are used as nodes and semantic relationships as edges to construct graph-structured data.

[0098] A graph consists of two elements: nodes and relationships. Each node represents an entity (a person, place, thing, category, or other data), and each relationship represents how two nodes are connected. This general structure can model a wide variety of scenarios—from road systems to networks of equipment to a population's medical history, or anything else defined by relationships. Graph databases store and query data using this data structure.

[0099] In the graph database for storing key entities, key terms, and semantic relationships in this embodiment, key entities are used as nodes and semantic relationships are used as edges to construct graph structured data and realize the storage of structured information.

[0100] Furthermore, a target index may be generated based on key entity names, term types, and semantic relationship types.

[0101] Among them, the target index is a multi-dimensional joint index, which supports compound queries based on entity attributes, clause content and relationship paths.

[0102] The method provided in this embodiment maps key entities to entity table fields in the relational database, and maps semantic relationships to foreign key constraints in the associated table, if the target database is a relational database. If the target database is a graph database, key entities are used as nodes and semantic relationships as edges to construct graph-structured data. Furthermore, a target index is generated based on the key entity name, clause type, and semantic relationship type. The target index is a multi-dimensional joint index that supports complex queries based on entity attributes, clause content, and relationship paths. This structured storage of contract text facilitates data mining and analysis, improving contract management effectiveness.

[0103] Optionally, the method further includes: Parse non-text elements in PDF or scanned contracts, including table content extraction and seal area recognition, and associate the parsing results with text structured data for storage.

[0104] Optionally, the method further includes: Monitor legal clause revisions and, when a revision is detected, crawl updated clause templates from authoritative legal databases; The updated term templates are injected into the rule matching library through the incremental learning algorithm, and the parameters of the NER model and SRL model are adjusted.

[0105] Figure 2 This is the second flow chart of the structured processing method of the contract text provided by the present invention, such as Figure 2As shown, the method includes the following steps: Step 201: Upload the contract text.

[0106] Step 202: Create classified contract items, including different types of contracts.

[0107] Step 203: Contract structuring processing.

[0108] Step 204 : Perform automatic verification in combination with the preset configuration of step 206 .

[0109] Step 205: After verification, the contract data elements are generated.

[0110] Step 206: Preset configuration: including format configurations such as the contract parties, contract date, contract amount, and product name.

[0111] The following describes the structured processing device for contract texts provided by the present invention. The structured processing device for contract texts described below and the structured processing method for contract texts described above can be referenced to each other.

[0112] Figure 3 This is a schematic diagram of the structure of the contract text structured processing device provided by the present invention. Figure 3 As shown, the contract text structured processing device 300 includes the following modules: A preprocessing module 310 is used to preprocess the contract text to generate a standardized word sequence; the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; An information extraction module 320 is configured to input the standardized word sequence into a named entity recognition model to identify key entities in the contract text and extract key terms in the contract text based on a machine learning algorithm; the key entities include the contracting parties, the contract amount, and the product name; and the key terms include payment terms, delivery terms, breach of contract liability terms, and force majeure terms. Semantic analysis module 330, configured to analyze the key entities and key terms using dependency syntax, and input the analysis results into a semantic role labeling model to obtain semantic relationships between the key entities and the key terms; the dependency syntax is used to identify grammatical structures between word sequences; The storage module 340 is configured to store the extracted key entities, key terms, and semantic relationships in a structured form in a target database, and to establish a target index to support fast query.

[0113] The device provided in this embodiment, first, a preprocessing module 310, is used to preprocess the contract text and generate a standardized word sequence, wherein the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; then, an information extraction module 320 is used to input the standardized word sequence into a named entity recognition model, identify key entities in the contract text, and extract key terms in the contract text based on a machine learning algorithm, wherein the key entities include the two parties to the contract, the contract amount and the product name, and the key terms include payment terms, delivery terms, breach of contract liability terms and force majeure terms; further, a semantic analysis module 330 is used to parse the key entities and the semantic relationships between the key terms through dependency syntax analysis and semantic role labeling model; and then, a storage module 340 is used to store the extracted key entities, key terms and semantic relationships in a structured form in a target database, and establish a target index to support fast query.

[0114] The present invention uses automated text processing and information extraction technology, and utilizes pre-trained models and machine learning algorithms to extract key information, obtain extracted key entities and key clauses, and use semantic role labeling models to parse the semantic relationships between key entities and key clauses. It can accurately identify key information and semantic relationships in the contract, and store the key information and semantic relationships in a structured manner, reducing the error rate of manual processing and thereby improving the efficiency of contract management and analysis.

[0115] According to a structured processing device 300 for a contract text provided by the present invention, the pre-processing module 310 is specifically configured to: removing special characters, redundant spaces, and line breaks from the contract text to obtain a cleaned contract text; Performing word segmentation processing on the cleaned contract text using an optimized word segmentation algorithm based on a contract domain dictionary to obtain multiple word sequences; Part-of-speech tagging is performed on each of the word sequences to obtain part-of-speech tagging information of each word sequence; the part-of-speech tagging information includes nouns, verbs, and adjectives.

[0116] According to a structured processing device 300 for a contract text provided by the present invention, the information extraction module 320 is specifically used to Performing character-level encoding on each of the word sequences to generate an input tensor corresponding to each of the word sequences including a word vector, a positional encoding, and a segment encoding; Inputting each of the input tensors into the named entity recognition model to extract each contextual semantic feature; Outputting entity label probability distribution information for each of the word sequences based on each of the contextual semantic features, and determining a target label sequence corresponding to each of the word sequences based on the entity label probability distribution information for each of the word sequences; the target label sequence is used to characterize the entity type corresponding to the word sequence; Continuous entity segments are merged according to target tag sequences corresponding to the word sequences to generate a structured entity list, and the structured entity list is determined as the identified key entities.

[0117] According to the structured processing device 300 for a contract text provided by the present invention, the information extraction module 320 is further configured to: Verify the contract domain rules on the identified structured entity list; If there are multiple candidate results for the same entity, ambiguity resolution is performed based on context confidence and rule priority.

[0118] According to the structured processing device 300 for a contract text provided by the present invention, the information extraction module 320 is further configured to: Based on a preset keyword library of terms, a bidirectional long short-term memory network model is used to classify and extract the terms to obtain the payment terms, the delivery terms, the breach of contract liability terms and the force majeure terms; The extracted key terms are subjected to confidence check. If the confidence is lower than the threshold, secondary matching and correction are performed in combination with the rule template.

[0119] According to the structured processing device 300 of a contract text provided by the present invention, the semantic analysis module 330 is specifically configured to: Performing dependency syntax analysis on the sentence containing the key entity to generate a syntax tree to identify the modification relationship between each of the word sequences; Inputting the syntax tree into the semantic role labeling model to obtain the semantic role of each word sequence; the semantic role of each word sequence is used to determine the subject, action, object relationship, condition and result relationship; According to the semantic roles of each of the word sequences, the contract subject layer, the obligation clause layer, the breach of contract clause layer and the force majeure clause layer are connected through directed edges to form a hierarchical association network.

[0120] According to the structured processing device 300 of a contract text provided by the present invention, the storage module 340 is specifically configured to: If the target database is a relational database, the key entity is mapped to an entity table field in the relational database, and the semantic relationship is mapped to a foreign key constraint of an association table; If the target database is a graph database, the key entities are used as nodes and the semantic relationships are used as edges to construct graph structure data; The target index is generated based on key entity names, clause types and semantic relationship types; the target index is a multi-dimensional joint index, and the target index supports compound queries based on entity attributes, clause content and relationship paths.

[0121] Figure 4 An example of a physical structure diagram of an electronic device is shown below. Figure 4 As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communication bus 440. The processor 410, the communications interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 may call logic instructions in the memory 430 to execute a structured contract text processing method, which includes: Preprocessing the contract text to generate a standardized word sequence; the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; Inputting the standardized word sequence into a named entity recognition model to identify key entities in the contract text, and extracting key terms in the contract text based on a machine learning algorithm; the key entities include the contracting parties, the contract amount, and the product name; and the key terms include payment terms, delivery terms, breach of contract liability terms, and force majeure terms; Analyzing the key entity and the key clause by using dependency syntax, and inputting the analysis results into a semantic role labeling model to obtain a semantic relationship between the key entity and the key clause; the dependency syntax is used to identify the grammatical structure between word sequences; The extracted key entities, key terms, and semantic relationships are stored in a structured form in a target database, and a target index is established to support fast query.

[0122] Furthermore, the logic instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0123] In another aspect, the present invention further provides a computer program product, comprising a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the structured processing method for a contract document provided by each of the above methods, the method comprising: Preprocessing the contract text to generate a standardized word sequence; the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; Inputting the standardized word sequence into a named entity recognition model to identify key entities in the contract text, and extracting key terms in the contract text based on a machine learning algorithm; the key entities include the contracting parties, the contract amount, and the product name; and the key terms include payment terms, delivery terms, breach of contract liability terms, and force majeure terms; Analyzing the key entity and the key clause by using dependency syntax, and inputting the analysis results into a semantic role labeling model to obtain a semantic relationship between the key entity and the key clause; the dependency syntax is used to identify the grammatical structure between word sequences; The extracted key entities, key terms, and semantic relationships are stored in a structured form in a target database, and a target index is established to support fast query.

[0124] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the structured processing method for the contract text provided by the above methods, comprising: Preprocessing the contract text to generate a standardized word sequence; the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; Inputting the standardized word sequence into a named entity recognition model to identify key entities in the contract text, and extracting key terms in the contract text based on a machine learning algorithm; the key entities include the contracting parties, the contract amount, and the product name; and the key terms include payment terms, delivery terms, breach of contract liability terms, and force majeure terms; Analyzing the key entity and the key clause by using dependency syntax, and inputting the analysis results into a semantic role labeling model to obtain a semantic relationship between the key entity and the key clause; the dependency syntax is used to identify the grammatical structure between word sequences; The extracted key entities, key terms, and semantic relationships are stored in a structured form in a target database, and a target index is established to support fast query.

[0125] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0126] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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 various embodiments of the present invention.

Claims

1. A method for structured processing of contract text, characterized in that: include: Preprocessing the contract text to generate a standardized word sequence; the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; Inputting the standardized word sequence into a named entity recognition model to identify key entities in the contract text, and extracting key terms in the contract text based on a machine learning algorithm; the key entities include the contracting parties, the contract amount, and the product name; and the key terms include payment terms, delivery terms, breach of contract liability terms, and force majeure terms; Analyzing the key entity and the key clause by using dependency syntax, and inputting the analysis results into a semantic role labeling model to obtain a semantic relationship between the key entity and the key clause; the dependency syntax is used to identify the grammatical structure between word sequences; The extracted key entities, key terms, and semantic relationships are stored in a structured form in a target database, and a target index is established to support fast query.

2. The method for structuring a contract text according to claim 1, characterized in that: The preprocessing of the contract text to generate a standardized word sequence includes: removing special characters, redundant spaces, and line breaks from the contract text to obtain a cleaned contract text; Performing word segmentation processing on the cleaned contract text using an optimized word segmentation algorithm based on a contract domain dictionary to obtain multiple word sequences; Part-of-speech tagging is performed on each of the word sequences to obtain part-of-speech tagging information of each word sequence; the part-of-speech tagging information includes nouns, verbs, and adjectives.

3. The method for structuring a contract text according to claim 1, characterized in that: Inputting the standardized word sequence into a named entity recognition model to identify key entities in the contract text includes: Performing character-level encoding on each of the word sequences to generate an input tensor corresponding to each of the word sequences including a word vector, a positional encoding, and a segment encoding; Inputting each of the input tensors into the named entity recognition model to extract each contextual semantic feature; Outputting entity label probability distribution information for each of the word sequences based on each of the contextual semantic features, and determining a target label sequence corresponding to each of the word sequences based on the entity label probability distribution information for each of the word sequences; the target label sequence is used to characterize the entity type corresponding to the word sequence; Continuous entity segments are merged according to target tag sequences corresponding to the word sequences to generate a structured entity list, and the structured entity list is determined as the identified key entities.

4. The method for structuring a contract text according to claim 3, characterized in that: After inputting the standardized word sequence into a named entity recognition model to identify key entities in the contract text, the method further includes: Verify the contract domain rules on the identified structured entity list; If there are multiple candidate results for the same entity, ambiguity resolution is performed based on context confidence and rule priority.

5. The method for structuring a contract text according to claim 1, characterized in that: The key terms extracted from the contract text based on the machine learning algorithm include: Based on a preset keyword library of terms, a bidirectional long short-term memory network model is used to classify and extract the terms to obtain the payment terms, the delivery terms, the breach of contract liability terms and the force majeure terms; The extracted key terms are subjected to confidence check. If the confidence is lower than the threshold, secondary matching and correction are performed in combination with the rule template.

6. The method for structuring a contract text according to claim 1, characterized in that: The step of analyzing the key entities and the key terms through dependency syntax and inputting the analysis results into a semantic role labeling model to obtain the semantic relationship between the key entities and the key terms includes: Performing dependency syntax analysis on the sentence containing the key entity to generate a syntax tree to identify the modification relationship between each of the word sequences; Inputting the syntax tree into the semantic role labeling model to obtain the semantic role of each word sequence; the semantic role of each word sequence is used to determine the subject, action, object relationship, condition and result relationship; According to the semantic roles of each of the word sequences, the contract subject layer, the obligation clause layer, the breach of contract clause layer and the force majeure clause layer are connected through directed edges to form a hierarchical association network.

7. The method for structuring a contract text according to claim 1, characterized in that: The step of storing the extracted key entities, key terms, and semantic relationships in a structured form in a target database and establishing a target index to support fast query includes: If the target database is a relational database, the key entity is mapped to an entity table field in the relational database, and the semantic relationship is mapped to a foreign key constraint of an association table; If the target database is a graph database, the key entities are used as nodes and the semantic relationships are used as edges to construct graph structure data; The target index is generated based on key entity names, clause types and semantic relationship types; the target index is a multi-dimensional joint index, and the target index supports compound queries based on entity attributes, clause content and relationship paths.

8. A structured processing device for a contract text, characterized in that: include: A preprocessing module is used to preprocess the contract text to generate a standardized word sequence; the standardized word sequence includes multiple word sequences and part-of-speech tagging information of each word sequence; an information extraction module, configured to input the standardized word sequence into a named entity recognition model, identify key entities in the contract text, and extract key terms in the contract text based on a machine learning algorithm; the key entities include the contracting parties, the contract amount, and the product name; and the key terms include payment terms, delivery terms, breach of contract liability terms, and force majeure terms; A semantic analysis module, configured to analyze the key entities and the key terms using dependency syntax, and input the analysis results into a semantic role labeling model to obtain the semantic relationship between the key entities and the key terms; the dependency syntax is used to identify the grammatical structure between word sequences; A storage module is used to store the extracted key entities, key terms and semantic relationships in a structured form in a target database, and to establish a target index to support fast query.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for structured processing of the contract text as claimed in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the structured processing method of the contract text as claimed in any one of claims 1 to 7.

Images