Scholarlink

This article describes the process of developing a module designed for automatic generation of patent application texts based on structured JSON data. The data is presented as a JSON file that contains annotated elements (tokens) and the relationships between them (relations). The main objective of the module is to transform the component structure of the invention into a coherent, correct text corresponding to the format of the description of the invention adopted in the patent documentation. The module uses a heuristic generator, graph structure analysis algorithms, and transformer-based language models (GPT-4o, Qwen2.5, and DeepSeek). The development was carried out in the Python programming language using the library pydantic (for typing and structuring data from JSON), network (for building component structure graphs), graphviz (for visualizing the structure of the invention), requests (for interacting with local language models (Qwen, DeepSeek)), g4f (for integration with GPT-4 through free APIs), sentence-transformers (for comparing texts by semantic similarity).

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development of a Neural Network Model for Generating the Text of a Patent Application

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sergey A. Fomenkov,
  • Svetlana A. Kozina,
  • Yaroslav D. Korobkin

摘要

This article describes the process of developing a module designed for automatic generation of patent application texts based on structured JSON data. The data is presented as a JSON file that contains annotated elements (tokens) and the relationships between them (relations). The main objective of the module is to transform the component structure of the invention into a coherent, correct text corresponding to the format of the description of the invention adopted in the patent documentation. The module uses a heuristic generator, graph structure analysis algorithms, and transformer-based language models (GPT-4o, Qwen2.5, and DeepSeek). The development was carried out in the Python programming language using the library pydantic (for typing and structuring data from JSON), network (for building component structure graphs), graphviz (for visualizing the structure of the invention), requests (for interacting with local language models (Qwen, DeepSeek)), g4f (for integration with GPT-4 through free APIs), sentence-transformers (for comparing texts by semantic similarity).