Research on Knowledge Extraction Based on Semantic Similarity
Hongjun Liu, Huili Gong, Xiangqian Ding · 2022 IEEE 6th Information Technology and Mechatronics Engineering Conference (ITOEC) · 2022
To solve the thorny problem of ‘information explosion’ in tunnel construction design, the knowledge extraction model based on semantic similarity (KEMSS) is proposed in this paper. In this model, the key words of designer's demand and the tunnel geological exploration and bidding texts are preprocessed; secondly, the Bert pretraining model is used to represent the two as semantic vectors containing context information; then the two feature vectors interact in the interaction layer to extract the matching features between texts; finally, semantic similarity is calculated by matching layer to extract the target text of the tunnel geological exploration and bidding texts. KEMSS is proposed for the semantic matching between short text and long text from a new perspective in this paper, which only needs a small part of the corpus in the field of tunnel construction to fine-tune based on knowledge transfer. The comparative experiments show that the accuracy and range of knowledge extraction of KEMSS are better than those of the comparative baseline models, which verifies the feasibility and effectiveness of the proposed model.