Short Text Matching Based on Multi-granularity Semantic Information

With the rapid development of online social networks, short text has become the main medium for users to transmit information. Chinese short text matching usually uses a sentence pair as output to obtain better results. However, there may be errors, ambiguities or inconsistencies in Chinese word segmentation, which affect the final laboratory matching effect. In order to solve this problem, we use the word lattice graph as input to retain multi-granularity information, so that the semantic information can be fully utilized. Experimental results on two Chinese data sets show that multi-granularity information improves text matching to some extent.