基于主题模型的绿色降碳技术知识图谱构建方法研究OA

Addressing the research bottlenecks in knowledge graph construction,such as the difficulty in integrating multi-source heterogeneous data and insufficient semantic mining,this paper proposes a knowledge graph construction method based on the vector-based topic model(VTM).Firstly,a semantic similarity optimization algorithm that incorporates context embeddings from the bidirectional encoder representations from Transformers(BERT)model at the sentence level is designed,achieving semantic disambiguation in terminology normalization.Secondly,a joint extraction mechanism for topic-entity relationships is constructed,enhancing the domain adaptability of entity recognition through dynamic topic constraints.Finally,a conflict resolution algorithm for multi-source entity alignment is designed,and a confidence evaluation model based on semantic vectors is established.When applying the proposed method to the construction of a knowledge graph for green carbon reduction technologies,experimental results show that,compared to existing BERT-based methods,the harmonic mean of precision and recall in named entity recognition and relation extraction tasks is improved by an average of 12.71%and 12.08%,respectively.