LeCAR: Leveraging Context for Enhanced Automotive Specification Retrieval

Abstract

In the domain of automotive manufacturing, specification documents represent intricate descriptions detailing every aspect of a product, design, or service. Conventionally, these specifications demand the deployment of expert teams to manually identify crucial data from the extensive documentation. The need to automate the extraction of candidate information from these documents is increasingly pressing in this industry. This research encounters two central challenges: Firstly, the queries for the specifications input by users are typically concise and ambiguous; secondly, not every word in a query carries the same significance. In response to these challenges, we propose LeCAR, which exploits contextual data to clarify query sentences and concentrate the search scope. Our experiments validate that the proposed method outperforms existing techniques that employ pre-trained language models, all without necessitating additional training data.