基于科技文献的锡铟材料知识图谱构建研究

Research on the Construction of Tin Indium Material Knowledge Map Based on Scientific and Technical Literature

随着科技文献数量的爆发性增长,垂直领域的企业面临着知识服务的挑战。为了帮助新材料领域企业有效利用科技文献信息资源,迫切需要使用人工智能技术对包含的材料性能等关键知识的专利及学术论文进行深入的知识建模,提供知识获取效率和准确性。以锡铟贵金属领域的10.48万篇科技文献为数据源,构建领域知识本体,利用BERT+BiLSTM+CRF模型进行命名实体识别,利用BERT+BiGRU神经网络模型进行关系抽取,将抽取结果存入图数据库Neo4j中并构建锡铟贵金属材料领域的知识图谱。所构建的知识图谱拥有18.19万个实体节点和23.47万条关系,能够实现多粒度的材料实体和关系的关联查询与可视化。基于锡铟科技文献构建知识图谱的方法拓展了新材料领域知识图谱构建的研究思路,为开展基于科技文献的垂直领域智能知识问答等应用提供了坚实的基础。

With the explosive growth in the volume of scientific literature,enterprises in vertical fields are facing challenges in knowledge services.To assist new material enterprises in effectively utilizing scientific and technical literature information resources,there is an urgent need to use artificial intelligence technology for in-depth knowledge modeling of patents and academic papers containing key knowledge,such as material properties,to enhance the efficiency and accuracy of knowledge acquisition.This study takes 104800 scientific and technical documents in the field of tin-indium precious metals as the data source,constructs a domain knowledge ontology,and employs the BERT+BiLSTM+CRF model for named entity recognition and the BERT+BiGRU neural network model for relationship extraction.The extracted results are stored in the graph database Neo4j,and a knowledge graph of the tin-indium precious metal material field is constructed.The resulting knowledge graph consists of 181900 entity nodes and 234700 relationships,enabling associated queries and visualization of material entities and relationships at multi-dimension.The method proposed in this paper for constructing a knowledge graph based on tin-indium scientific and technical literature not only expands the research approach for building knowledge graph in the new material field but also provides a solid foundation for the development of vertical field intelligent knowledge Q&A applications based on scientific and technical literature.