Multi-granularity semantic collaborative modeling method for new energy vehicle fault description and related equipment

Abstract

This application discloses a multi-granularity semantic collaborative modeling method and related equipment for fault description in new energy vehicles, which can be applied to the fields of intelligent vehicle diagnosis and natural language processing. This application obtains an input sequence matrix by performing domain-adaptive input encoding on the original fault description text based on a pre-built new energy vehicle domain dictionary. It then performs contextual semantic modeling on the original fault description text to obtain a contextual semantic representation matrix. Next, it extracts features and calculates weights from the contextual semantic representation matrix to obtain a global intent representation vector. After obtaining a preliminary semantic parsing representation through joint encoding of intent and slots using a pre-trained intent classification module and slot annotation module, it converts the preliminary semantic parsing representation into a structured semantic parsing representation. This enables high-precision, fine-grained intent recognition from complex descriptions, improves the accuracy of semantic slot extraction, and effectively supports downstream causal reasoning for new energy vehicle faults.

Images