基于深度学习的可降解农用地膜领域专利知识图谱构建与应用

• 作者： 胡智杰 ^1,2 何莹 ^1,2 吴俣 ^1,2 孙巍 ^1,2 马晓敏 ^1,2 张学福 ^1,2
• 作者单位：
  1.  中国农业科学院农业信息研究所

  2.  农业农村部农业大数据重点实验室
• 提交时间：2024-09-11

摘要: 为解决可降解农用地膜领域知识结构复杂、专利文本专有名词众多、数据利用效率低等问题，本文利用知识图谱强大的语义处理和知识关联能力，提出了一种基于深度学习的可降解农用地膜领域专利知识图谱构建方法。首先，该方法根据可降解农用地膜专利数据特征并结合专家先验知识，预定义实体类型、属性和关系集合，采用自顶向下的模式构建知识图谱本体模型。随后，按照本体模型选定数据范围，从结构化、半结构化和非结构化专利数据中进行数据选取或信息抽取。对于半结构化数据和非结构化数据部分，对少量训练样本进行文本标注后，利用通用信息抽取（Universal information extraction，UIE）框架进行数据挖掘，识别并抽取可降解农用地膜技术要素字段。抽取结果表明，在少量训练样本前提下，该抽取方法具有较高的准确率97.78%，召回率93.62%和F1值95.65%，显著优于BERT-BiLSTM+CRF模型。最后，将抽取到的知识存储到Neo4j图数据库，实现知识图谱可视化及知识关联推理。本研究构建的可降解农用地膜知识图谱可以有效解析可降解农用地膜领域技术结构，实现对农膜专利数据的有效分析挖掘，并可较好的为地膜专利信息检索、知识问答等应用领域提供知识库基础。

Abstract: Aiming at the problems of complex knowledge structure, numerous patent proper nouns and low data utilization efficiency in the field of degradable agricultural mulching film, this paper proposes a method of constructing patent knowledge map in the field of degradable agricultural mulching film based on deep learning by utilizing the powerful semantic processing and knowledge association capabilities of knowledge map. Firstly, according to the characteristics of degradable agricultural mulching film patent data and combined with expert prior knowledge, the entity types, attributes and relationship sets were predefined, and the top-down mode was used to construct the knowledge graph ontology model. Then, the scope of data is selected according to the ontology model, and data selection or information extraction is carried out from structured, semi-structured and unstructured patent data. For the semi-structured data and unstructured data, a small number of training samples were text labeled, and the Universal information extraction (UIE) framework was used for data mining to identify and extract degradable agricultural mulching film technical element fields. The extraction results show that under the premise of a small number of training samples, the extraction method has a high accuracy of 97.78%, recall rate of 93.62% and F1 value of 95.65%, which is significantly better than the BERT-BiLSTM+CRF model. Finally, the extracted knowledge was stored in the Neo4j graph database to realize knowledge graph visualization and knowledge association reasoning. The knowledge graph of degradable agricultural mulching film constructed in this study can effectively analyze the technical structure of degradable agricultural mulching film, realize the effective analysis and mining of agricultural mulching film patent data, and provide a knowledge base basis for application fields such as information retrieval and knowledge question answering.

• 来自： 胡智杰
• 分类： 信息资源管理 >> 情报学
• 稿件状态： 已投稿会议
会议名称： 数智赋能学术期刊创新发展 ——第四届信息资源管理期刊发展论坛征文（2024年7月 —2024年10月）
• 引用： PSSXiv:202409.00694 (或此版本 PSSXiv:202409.00694V1)
  DOI:10.12451/202409.00694
  CSTR:32012.36.PSSXiv.202409.00694
• 推荐引用方式： 胡智杰,何莹,吴俣,孙巍,马晓敏,张学福.基于深度学习的可降解农用地膜领域专利知识图谱构建与应用.哲学社会科学预印本平台:https://zsyyb.cn/abs/202409.00694.[PSSXiv:202409.00694V1] (点此复制)