Short Text Matching Enhanced by Contextual Template for Large Language Models in Data Exchanging Systems

Large language models (LLMs) have demonstrated remarkable capabilities in understanding and matching texts of moderate length. However, their performance in short text matching falls short of expectations. This paper proposes a novel approach for LLM-based short text matching, Contextual Template Enhancement (CTE), to improve short text matching accuracy by leveraging contextual information. CTE enhances short texts by incorporating contextual details into templates, transforming short texts into longer texts that better preserve the original semantics while enriching their representational power. Experiments in a data exchanging system show that CTE significantly improves the matching accuracy between short data annotations and data requests.