LET: Linguistic Knowledge Enhanced Graph Transformer for Chinese Short Text Matching

Chinese short text matching is a fundamental task in natural language processing. Existing approaches usually take characters or words as input tokens. They have two limitations: 1) Some are polysemous, and semantic information not fully utilized. 2) models suffer potential issues caused by word segmentation. Here we introduce HowNet an external knowledge base propose Linguistic Enhanced graph Transformer (LET) to deal with ambiguity. Additionally, adopt the lattice maintain multi-granularity information. Our model also complementary pre-trained models. Experimental results on datasets show that our outperform various typical approaches. Ablation study indicates both important for modeling.