Intelligent document generation method based on Java

Abstract

The invention discloses a Java-based intelligent document generation method, and belongs to the technical field of intelligent generation, and the method comprises the steps: S1, constructing a deep learning model: constructing a natural language processing model based on large-scale legal text data set training by using a deep learning framework integrated by a Java platform, the natural language processing model is used for understanding complex legal concepts, terms and logic relations, in the method, deep law understanding ability is achieved, deep understanding of legal texts is achieved through application of the deep learning model, a shallow processing mode of traditional template filling is exceeded, efficiency is improved, the legal document preparation time is greatly shortened, and the method is suitable for popularization and application. The workload of legal professionals is reduced, the accuracy is enhanced, document errors and omissions are reduced, the specialty and compliance of legal documents are ensured, standardization is promoted, the standardized process of legal document writing is pushed, the consistency and quality of legal services are improved, personalized services are achieved, and the special requirements of specific clients or cases are met.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent generation, and in particular to a Java-based intelligent document generation method. Background Art

[0002] In the current information age, the legal industry is also experiencing a wave of digital transformation. Traditional legal document writing relies on manual writing by lawyers or legal staff. This process is not only time-consuming and labor-intensive, but also prone to errors and omissions due to human factors. In recent years, with the rapid development of natural language processing (NLP) technology, especially intelligent applications on the Java platform, it has become possible for the automated generation of legal documents.

[0003] The prior art discloses some invention patents in the technical field of intelligent generation, among which the invention patent with publication number CN103412868A discloses a document generation method and device. By configuring two data tables, no matter what document the user selects or adds a new document type, the corresponding code used in the program does not need to be changed. It is only necessary to configure one or two of the data tables to implement operations including adding new document templates and changing the content to be generated in the document template, which is more convenient for later use.

[0004] Existing technologies attempt to automatically generate basic legal documents through methods such as template filling and keyword recognition. However, these systems generally lack flexibility, have difficulty adapting to complex and changing legal situations, and lack the ability to deeply understand and customize. Therefore, the market urgently needs a document generation technology that is more intelligent, accurate, and can deeply understand legal logic. Summary of the invention

[0005] The purpose of the present invention is to propose a Java-based intelligent document generation method to solve the problems that existing technologies attempt to automatically generate basic legal documents through methods such as template filling and keyword recognition, but these systems generally have insufficient flexibility, difficulty in adapting to complex and changing legal situations, and lack of deep understanding and personalized customization capabilities.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A Java-based intelligent document generation method comprises the following steps:

[0008] S1: Build a deep learning model, using the deep learning framework integrated in the Java platform to build a natural language processing model trained on a large-scale legal text dataset. The natural language processing model is used to understand complex legal concepts, clauses, and logical relationships.

[0009] S2: Integrate the legal knowledge graph to build a knowledge graph containing rich legal provisions, cases, terminology and logical rules, and seamlessly connect it with the intelligent generation system through the Java application program interface;

[0010] S3: Develop a personalized template engine. Develop a flexible template engine that allows users to define text structure and style according to different legal requirements. The engine is combined with Java's powerful programming capabilities and template language to achieve dynamic content filling and logical reasoning;

[0011] S4: Interactive editing and review: Design a user-friendly interface using Java Swing or JavaFX graphics library to enable users to quickly modify and review the draft after it is generated. The system provides trial feedback and suggestions to assist users in improving the document.

[0012] As a further description of the above technical solution:

[0013] The step S2 further comprises:

[0014] S21: Use the structural information of the legal text dataset to guide the generation of legal documents, convert the legal document generation task into multiple subtasks, solve the long text encoding problem, and enable the natural language processing model to focus on the core content of the original legal text;

[0015] S22: A maximum similarity matching method based on internal knowledge is used to process legal text noise using the correlation between subtasks, solve the problem of information redundancy, help the natural language processing model obtain deeper semantic information, and further improve the quality of legal documents;

[0016] S23: A legal document generation method is proposed that is guided by structural information and integrates internal knowledge, so that the natural language processing model can quickly capture the key information of the legal text.

[0017] As a further description of the above technical solution:

[0018] The deep learning framework is TensorFlow or PyTorch for Java;

[0019] The large-scale legal text data includes government public data sources, text databases, journals, web crawler technology, partners and data sharing, crowdsourcing and volunteer contributions, official publications and reports, copyright licenses and purchases, social media and forums, legal education materials and case studies. The government public data sources include legal provisions, court decisions and legislative records that are made public by national and regional government agencies. The government public data sources are obtained from official legal databases, court websites and bulletin boards of legislative bodies.

[0020] As a further description of the above technical solution:

[0021] The method for constructing the natural language processing model based on the Java platform includes:

[0022] Data preprocessing:

[0023] Data collection, collecting text data, including social media posts, news articles, and legal documents;

[0024] Data cleaning: clean the data, remove HTML tags, punctuation marks, and stop words, and retain meaningful information;

[0025] Data conversion: converting text data into a format suitable for machine learning models, using Java string processing functions or third-party libraries;

[0026] Feature extraction:

[0027] Lexical analysis, using Java libraries for part-of-speech tagging and named entity recognition, such as StanfordCoreNLP, OpenNLP, or JWNL;

[0028] Vectorization, which converts text into numerical representations, is implemented using TF-IDF vectors, word embeddings (Word2Vec and GloVe), Deeplearning4j libraries, or Encog libraries;

[0029] Model training:

[0030] Select a model, and select an appropriate model according to the task type, the task type includes classification, sequence labeling and generation, and the model includes convolutional neural network (CNN), long short-term neural network (LSTM) and Transformer;

[0031] Train the model using the collected dataset, based on the deep learning framework Deeplearning4j (DL4J) on the Java platform, using the Java API to train the model, or use Python libraries (TensorFlow or PyTorch) and interact with the Java environment through Jython or Jep type bridge tools;

[0032] Model Evaluation:

[0033] Validation and testing, using cross-validation techniques to evaluate model performance and ensure that the model has strong generalization capabilities;

[0034] Metrics calculation, calculating the accuracy, recall and F1 score evaluation metrics to determine the effectiveness of the model;

[0035] Model deployment:

[0036] Encapsulate the model and encapsulate the trained model into a Java application so that it can be called by other systems;

[0037] API development, creating RESTful APIs or other forms of interfaces for external calls to models for prediction or analysis.

[0038] As a further description of the above technical solution:

[0039] The Stanford CoreNLP is used to provide a series of NLP tools, including syntactic analysis, semantic role labeling and sentiment analysis;

[0040] The OpenNLP is used to process natural language texts and provide named entity recognition, word segmentation and part-of-speech tagging;

[0041] The Deeplearning4j (DL4J) is an open source distributed deep learning library for building and training neural network models;

[0042] Encog is a Java machine learning framework that supports neural networks and genetic algorithms.

[0043] As a further description of the above technical solution:

[0044] The Java application program and the intelligent generation system are seamlessly connected via an API interface.

[0045] As a further description of the above technical solution:

[0046] The construction of the knowledge graph includes:

[0047] Data collection, collecting structured data, semi-structured data and unstructured data, the structured data includes databases, XML files and CSV files, the semi-structured data includes HTML web pages and JSON files, and the unstructured data includes text files, PDF documents and social media posts;

[0048] Data preprocessing, cleaning and integrating the collected data, wherein the data cleaning includes removing noise, correcting errors and unifying the format, and the data integration includes merging duplicate information and resolving inconsistency issues;

[0049] Knowledge extraction, collecting entities, relations and attributes from data, the entity recognition (NER) includes automatically identifying entities of types such as people, places and organizations in text, the relationship extraction includes detecting the relationship between entities, where the relationship between entities includes work place and birthplace, the attribute extraction includes identifying the attributes of entities, where entity attributes include age and gender;

[0050] Knowledge representation: convert the extracted knowledge into the standard format of knowledge graph and represent it using RDF (Resource Description Framework) triple method;

[0051] Subject construction, defining the ontology of the knowledge graph, including the conceptual model describing entities, attributes, and relationships, to help ensure the consistency and scalability of the knowledge graph;

[0052] Knowledge fusion combines knowledge extracted from different sources to solve the problem of different representations of the same entity, ensuring the completeness and accuracy of the graph, where different representations include synonyms and aliases;

[0053] Knowledge verification: automatic verification and manual review of the extracted and integrated knowledge to ensure the correctness and credibility of the knowledge;

[0054] Knowledge graph storage: Choose a suitable storage solution, including RDF database and graph database, to store the knowledge graph for query and analysis;

[0055] Query and reasoning: developing query interfaces and reasoning engines to enable users or applications to query knowledge graphs and perform rule-based reasoning;

[0056] Update and maintain: regularly update the knowledge graph, add new knowledge, correct errors, and keep the graph timely and effective;

[0057] Construction method, the knowledge graph is constructed using top-down, bottom-up or a combination of the two methods. The top-down method starts from existing structured data and ontologies and gradually fills in the details. The bottom-up method extracts patterns from a large amount of unstructured data and gradually constructs ontologies and knowledge graphs.

[0058] As a further description of the above technical solution:

[0059] The development of the template engine includes:

[0060] Design the syntax, data model and architecture of the template engine. The template engine syntax includes determining the tags, variables and control structures used in the template. The control structures include if statements and for loops. The definition of the data model includes planning how the engine will receive and process data and completing the design of the data binding mechanism. The architecture decision includes choosing to develop an interpreted engine or a compiled engine. The interpreted engine directly parses and executes the template, while the compiled engine converts the template into intermediate code or target language code before executing it.

[0061] Select development technology and tools, programming language, select programming language Java;

[0062] Data bundling and rendering logic, which respectively perform data bundling, conditions and loops, and plug-ins and extensions. The data bundling is used to map the incoming data to the variables in the template. The conditions and loops are used to implement conditional statements and loop control structures to dynamically generate content based on the data. The plug-ins and extensions consider whether to allow plug-ins or custom tags to increase the flexibility of the template engine.

[0063] Documentation and user guide, complete writing documents and creating sample templates. The written documents are used to provide users with detailed documents on how to use the template engine, including syntax guides and examples. The created sample templates are used to provide some sample templates and data to help users get started quickly.

[0064] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0065] 1. In the present invention, the deep legal understanding ability, through the use of deep learning models, realizes a deep understanding of legal texts, surpasses the shallow processing method of traditional template filling, improves efficiency, greatly shortens the preparation time of legal documents, reduces the workload of legal professionals, enhances accuracy, reduces document errors and omissions, ensures the professionalism and compliance of legal documents, promotes standardization, promotes the standardization process of legal document writing, improves the consistency and quality of legal services, personalizes services, meets the special needs of specific clients or cases, and improves customer satisfaction with legal services.

[0066] 2. In the present invention, dynamic knowledge graph fusion, dynamically updated and queried legal knowledge graph ensures the timeliness and accuracy of generated documents.

[0067] 3. The present invention is highly customizable, and users can customize templates and logical rules to meet personalized needs in different scenarios, thereby improving the flexibility and practicality of the system.

[0068] 4. In the present invention, an interactive optimization experience is provided, an interactive editing interface and intelligent auxiliary review functions are provided, the workflow of legal workers is simplified, and the efficiency of document production is improved.

[0069] 5. In the present invention, the legal document generation method that integrates internal knowledge enables the natural language processing model to quickly capture the key information of the legal text, improve the convergence speed of the natural language processing model, generate legal documents with clear structure and rich content, and provide new ideas for long text tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1A schematic diagram of a flow chart of a Java-based intelligent document generation method proposed by the present invention;

[0071] Figure 2 A flowchart of a legal document generation method based on trial logic steps in a Java-based intelligent document generation method proposed by the present invention;

[0072] Figure 3 A flowchart of a method for constructing a natural language processing model based on a Java platform in a Java-based intelligent document generation method proposed by the present invention;

[0073] Figure 4 A schematic diagram of the process of constructing a knowledge graph in a Java-based intelligent document generation method proposed in the present invention;

[0074] Figure 5 The present invention is a flowchart of the development of a template engine in a Java-based intelligent document generation method. DETAILED DESCRIPTION

[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] Embodiment 1:

[0077] A Java-based intelligent document generation method comprises the following steps:

[0078] S1: Build a deep learning model, using the deep learning framework integrated in the Java platform to build a natural language processing model trained on a large-scale legal text dataset. The natural language processing model is used to understand complex legal concepts, clauses, and logical relationships.

[0079] S2: Integrate the legal knowledge graph to build a knowledge graph containing rich legal provisions, cases, terminology and logical rules, and seamlessly connect it with the intelligent generation system through the Java application program interface;

[0080] S3: Develop a personalized template engine. Develop a flexible template engine that allows users to define text structure and style according to different legal requirements. The engine is combined with Java's powerful programming capabilities and template language to achieve dynamic content filling and logical reasoning;

[0081] S4: Interactive editing and review: Design a user-friendly interface using Java Swing or JavaFX graphics library to enable users to quickly modify and review the draft after it is generated. The system provides trial feedback and suggestions to assist users in improving the document.

[0082] As a further description of the above technical solution:

[0083] The deep learning framework is TensorFlow or PyTorch for Java;

[0084] The large-scale legal text data includes government public data sources, text databases, journals, web crawler technology, partners and data sharing, crowdsourcing and volunteer contributions, official publications and reports, copyright licenses and purchases, social media and forums, legal education materials and case studies. The government public data sources include legal provisions, court decisions and legislative records that are made public by national and regional government agencies. The government public data sources are obtained from official legal databases, court websites and bulletin boards of legislative bodies.

[0085] As a further description of the above technical solution:

[0086] The method for constructing the natural language processing model based on the Java platform includes:

[0087] Data preprocessing:

[0088] Data collection, collecting text data, including social media posts, news articles, and legal documents;

[0089] Data cleaning: clean the data, remove HTML tags, punctuation marks, and stop words, and retain meaningful information;

[0090] Data conversion: converting text data into a format suitable for machine learning models, using Java string processing functions or third-party libraries;

[0091] Feature extraction:

[0092] Lexical analysis, using Java libraries for part-of-speech tagging and named entity recognition, such as StanfordCoreNLP, OpenNLP, or JWNL;

[0093] Vectorization, which converts text into numerical representations, is implemented using TF-IDF vectors, word embeddings (Word2Vec and GloVe), Deeplearning4j libraries, or Encog libraries;

[0094] Model training:

[0095] Select a model, and select an appropriate model according to the task type, the task type includes classification, sequence labeling and generation, and the model includes convolutional neural network (CNN), long short-term neural network (LSTM) and Transformer;

[0096] Train the model using the collected dataset, based on the deep learning framework Deeplearning4j (DL4J) on the Java platform, using the Java API to train the model, or use Python libraries (TensorFlow or PyTorch) and interact with the Java environment through Jython or Jep type bridge tools;

[0097] Model Evaluation:

[0098] Validation and testing, using cross-validation techniques to evaluate model performance and ensure that the model has strong generalization capabilities;

[0099] Metrics calculation, calculating the accuracy, recall and F1 score evaluation metrics to determine the effectiveness of the model;

[0100] Model deployment:

[0101] Encapsulate the model and encapsulate the trained model into a Java application so that it can be called by other systems;

[0102] API development, creating RESTful APIs or other forms of interfaces for external calls to models for prediction or analysis.

[0103] As a further description of the above technical solution:

[0104] The Stanford CoreNLP is used to provide a series of NLP tools, including syntactic analysis, semantic role labeling and sentiment analysis;

[0105] The OpenNLP is used to process natural language texts and provide named entity recognition, word segmentation and part-of-speech tagging;

[0106] The Deeplearning4j (DL4J) is an open source distributed deep learning library for building and training neural network models;

[0107] Encog is a Java machine learning framework that supports neural networks and genetic algorithms.

[0108] As a further description of the above technical solution:

[0109] The Java application program and the intelligent generation system are seamlessly connected via an API interface.

[0110] As a further description of the above technical solution:

[0111] The construction of the knowledge graph includes:

[0112] Data collection, collecting structured data, semi-structured data and unstructured data, the structured data includes databases, XML files and CSV files, the semi-structured data includes HTML web pages and JSON files, and the unstructured data includes text files, PDF documents and social media posts;

[0113] Data preprocessing, cleaning and integrating the collected data, wherein the data cleaning includes removing noise, correcting errors and unifying the format, and the data integration includes merging duplicate information and resolving inconsistency issues;

[0114] Knowledge extraction, collecting entities, relations and attributes from data, the entity recognition (NER) includes automatically identifying entities of types such as people, places and organizations in text, the relationship extraction includes detecting the relationship between entities, where the relationship between entities includes work place and birthplace, the attribute extraction includes identifying the attributes of entities, where entity attributes include age and gender;

[0115] Knowledge representation: convert the extracted knowledge into the standard format of knowledge graph and represent it using RDF (Resource Description Framework) triple method;

[0116] Subject construction, defining the ontology of the knowledge graph, including the conceptual model describing entities, attributes, and relationships, to help ensure the consistency and scalability of the knowledge graph;

[0117] Knowledge fusion combines knowledge extracted from different sources to solve the problem of different representations of the same entity, ensuring the completeness and accuracy of the graph, where different representations include synonyms and aliases;

[0118] Knowledge verification: automatic verification and manual review of the extracted and integrated knowledge to ensure the correctness and credibility of the knowledge;

[0119] Knowledge graph storage: Choose a suitable storage solution, including RDF database and graph database, to store the knowledge graph for query and analysis;

[0120] Query and reasoning: developing query interfaces and reasoning engines to enable users or applications to query knowledge graphs and perform rule-based reasoning;

[0121] Update and maintain: regularly update the knowledge graph, add new knowledge, correct errors, and keep the graph timely and effective;

[0122] Construction method, the knowledge graph is constructed using top-down, bottom-up or a combination of the two methods. The top-down method starts from existing structured data and ontologies and gradually fills in the details. The bottom-up method extracts patterns from a large amount of unstructured data and gradually constructs ontologies and knowledge graphs.

[0123] As a further description of the above technical solution:

[0124] The development of the template engine includes:

[0125] Design the syntax, data model and architecture of the template engine. The template engine syntax includes determining the tags, variables and control structures used in the template. The control structures include if statements and for loops. The definition of the data model includes planning how the engine will receive and process data and completing the design of the data binding mechanism. The architecture decision includes choosing to develop an interpreted engine or a compiled engine. The interpreted engine directly parses and executes the template, while the compiled engine converts the template into intermediate code or target language code before executing it.

[0126] Select development technology and tools, programming language, select programming language Java;

[0127] Data bundling and rendering logic, which respectively perform data bundling, conditions and loops, and plug-ins and extensions. The data bundling is used to map the incoming data to the variables in the template. The conditions and loops are used to implement conditional statements and loop control structures to dynamically generate content based on the data. The plug-ins and extensions consider whether to allow plug-ins or custom tags to increase the flexibility of the template engine.

[0128] Documentation and user guide, complete writing documents and creating sample templates. The written documents are used to provide users with detailed documents on how to use the template engine, including syntax guides and examples. The created sample templates are used to provide some sample templates and data to help users get started quickly.

[0129] Embodiment 2:

[0130] A Java-based intelligent document generation method comprises the following steps:

[0131] S1: Build a deep learning model, using the deep learning framework integrated in the Java platform to build a natural language processing model trained on a large-scale legal text dataset. The natural language processing model is used to understand complex legal concepts, clauses, and logical relationships.

[0132] S21: Use the structural information of the legal text dataset to guide the generation of legal documents, convert the legal document generation task into multiple subtasks, solve the long text encoding problem, and enable the natural language processing model to focus on the core content of the original legal text;

[0133] S22: A maximum similarity matching method based on internal knowledge is used to process legal text noise using the correlation between subtasks, solve the problem of information redundancy, help the natural language processing model obtain deeper semantic information, and further improve the quality of legal documents;

[0134] S23: A legal document generation method is proposed that is guided by structural information and integrates internal knowledge, so that the natural language processing model can quickly capture the key information of the legal text;

[0135] S3: Develop a personalized template engine. Develop a flexible template engine that allows users to define text structure and style according to different legal requirements. The engine is combined with Java's powerful programming capabilities and template language to achieve dynamic content filling and logical reasoning;

[0136] S4: Interactive editing and review: Design a user-friendly interface using Java Swing or JavaFX graphics library to enable users to quickly modify and review the draft after it is generated. The system provides trial feedback and suggestions to assist users in improving the document.

[0137] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A Java-based intelligent document generation method, characterized in that: Includes steps: S1: Build a deep learning model, using the deep learning framework integrated in the Java platform to build a natural language processing model trained on a large-scale legal text dataset. The natural language processing model is used to understand complex legal concepts, clauses, and logical relationships. S2: Integrate the legal knowledge graph to build a knowledge graph containing rich legal provisions, cases, terminology and logical rules, and seamlessly connect it with the intelligent generation system through the Java application program interface; S3: Develop a personalized template engine. Develop a flexible template engine that allows users to define text structure and style according to different legal requirements. The engine is combined with Java's powerful programming capabilities and template language to achieve dynamic content filling and logical reasoning; S4: Interactive editing and review: Design a user-friendly interface using Java Swing or JavaFX graphics library to enable users to quickly modify and review the draft after it is generated. The system provides trial feedback and suggestions to assist users in improving the document.

2. The Java-based intelligent document generation method according to claim 1, characterized in that: The step S2 also includes a method for generating legal documents based on the trial logic step: S21: Use the structural information of the legal text dataset to guide the generation of legal documents, convert the legal document generation task into multiple subtasks, solve the long text encoding problem, and enable the natural language processing model to focus on the core content of the original legal text; S22: A maximum similarity matching method based on internal knowledge is used to process legal text noise using the correlation between subtasks, solve the problem of information redundancy, help the natural language processing model obtain deeper semantic information, and further improve the quality of legal documents; S23: A legal document generation method is proposed that is guided by structural information and integrates internal knowledge, so that the natural language processing model can quickly capture the key information of the legal text.

3. The Java-based intelligent document generation method according to claim 1, characterized in that: The deep learning framework is TensorFlow or PyTorch for Java; The large-scale legal text data includes government public data sources, text databases, journals, web crawler technology, partners and data sharing, crowdsourcing and volunteer contributions, official publications and reports, copyright licenses and purchases, social media and forums, legal education materials and case studies. The government public data sources include legal provisions, court decisions and legislative records that are made public by national and regional government agencies. The government public data sources are obtained from official legal databases, court websites and bulletin boards of legislative bodies.

4. The Java-based intelligent document generation method according to claim 3 is characterized in that: The method for constructing the natural language processing model based on the Java platform includes: Data preprocessing: Data collection, collecting text data, including social media posts, news articles, and legal documents; Data cleaning: clean the data, remove HTML tags, punctuation marks, and stop words, and retain meaningful information; Data conversion: converting text data into a format suitable for machine learning models, using Java string processing functions or third-party libraries; Feature extraction: Lexical analysis, using Java libraries for part-of-speech tagging and named entity recognition, such as StanfordCoreNLP, OpenNLP, or JWNL; Vectorization, which converts text into numerical representations, is implemented using TF-IDF vectors, word embeddings (Word2Vec and GloVe), Deeplearning4j libraries, or Encog libraries; Model training: Select a model, and select an appropriate model according to the task type, the task type includes classification, sequence labeling and generation, and the model includes convolutional neural network (CNN), long short-term neural network (LSTM) and Transformer; Train the model using the collected dataset, based on the deep learning framework Deeplearning4j (DL4J) on the Java platform, using the Java API to train the model, or use Python libraries (TensorFlow or PyTorch) and interact with the Java environment through Jython or Jep type bridge tools; Model Evaluation: Validation and testing, using cross-validation techniques to evaluate model performance and ensure that the model has strong generalization capabilities; Metrics calculation, calculating the accuracy, recall and F1 score evaluation metrics to determine the effectiveness of the model; Model deployment: Encapsulate the model and encapsulate the trained model into a Java application so that it can be called by other systems; API development, creating RESTful APIs or other forms of interfaces for external calls to models for prediction or analysis.

5. The Java-based intelligent document generation method according to claim 4 is characterized in that: The Stanford CoreNLP is used to provide a series of NLP tools, including syntactic analysis, semantic role labeling and sentiment analysis; The OpenNLP is used to process natural language texts and provide named entity recognition, word segmentation and part-of-speech tagging; The Deeplearning4j (DL4J) is an open source distributed deep learning library for building and training neural network models; Encog is a Java machine learning framework that supports neural networks and genetic algorithms.

6. The Java-based intelligent document generation method according to claim 2, characterized in that: The Java application program and the intelligent generation system are seamlessly connected via an API interface.

7. The Java-based intelligent document generation method according to claim 6, characterized in that: The construction of the knowledge graph includes: Data collection, collecting structured data, semi-structured data and unstructured data, the structured data includes databases, XML files and CSV files, the semi-structured data includes HTML web pages and JSON files, and the unstructured data includes text files, PDF documents and social media posts; Data preprocessing, cleaning and integrating the collected data, wherein the data cleaning includes removing noise, correcting errors and unifying the format, and the data integration includes merging duplicate information and resolving inconsistency issues; Knowledge extraction, collecting entities, relations and attributes from data, the entity recognition (NER) includes automatically identifying entities of types such as people, places and organizations in text, the relationship extraction includes detecting the relationship between entities, where the relationship between entities includes work place and birthplace, the attribute extraction includes identifying the attributes of entities, where entity attributes include age and gender; Knowledge representation: convert the extracted knowledge into the standard format of knowledge graph and represent it using RDF (Resource Description Framework) triple method; Subject construction, defining the ontology of the knowledge graph, including the conceptual model describing entities, attributes, and relationships, to help ensure the consistency and scalability of the knowledge graph; Knowledge fusion combines knowledge extracted from different sources to solve the problem of different representations of the same entity, ensuring the completeness and accuracy of the graph, where different representations include synonyms and aliases; Knowledge verification: automatic verification and manual review of the extracted and integrated knowledge to ensure the correctness and credibility of the knowledge; Knowledge graph storage: Choose a suitable storage solution, including RDF database and graph database, to store the knowledge graph for query and analysis; Query and reasoning: developing query interfaces and reasoning engines to enable users or applications to query knowledge graphs and perform rule-based reasoning; Update and maintain: regularly update the knowledge graph, add new knowledge, correct errors, and keep the graph timely and effective; Construction method, the knowledge graph is constructed using top-down, bottom-up or a combination of the two methods. The top-down method starts from existing structured data and ontologies and gradually fills in the details. The bottom-up method extracts patterns from a large amount of unstructured data and gradually constructs ontologies and knowledge graphs.

8. The Java-based intelligent document generation method according to claim 1, characterized in that: The development of the template engine includes: Design the syntax, data model and architecture of the template engine. The template engine syntax includes determining the tags, variables and control structures used in the template. The control structures include if statements and for loops. The definition of the data model includes planning how the engine will receive and process data and completing the design of the data binding mechanism. The architecture decision includes choosing to develop an interpreted engine or a compiled engine. The interpreted engine directly parses and executes the template, while the compiled engine converts the template into intermediate code or target language code before executing it. Select development technology and tools, programming language, select programming language Java; Data bundling and rendering logic, which respectively perform data bundling, conditions and loops, and plug-ins and extensions. The data bundling is used to map the incoming data to the variables in the template. The conditions and loops are used to implement conditional statements and loop control structures to dynamically generate content based on the data. The plug-ins and extensions consider whether to allow plug-ins or custom tags to increase the flexibility of the template engine. Documentation and user guide, complete writing documents and creating sample templates. The written documents are used to provide users with detailed documents on how to use the template engine, including syntax guides and examples. The created sample templates are used to provide some sample templates and data to help users get started quickly.

Images