Abstract
This paper presents CLaSP, a novel model for retrieving time-series signalsusing natural language queries that describe signal characteristics. Theability to search time-series signals based on descriptive queries is essentialin domains such as industrial diagnostics, where data scientists often need tofind signals with specific characteristics. However, existing methods rely onsketch-based inputs, predefined synonym dictionaries, or domain-specific manualdesigns, limiting their scalability and adaptability. CLaSP addresses thesechallenges by employing contrastive learning to map time-series signals tonatural language descriptions. Unlike prior approaches, it eliminates the needfor predefined synonym dictionaries and leverages the rich contextual knowledgeof large language models (LLMs). Using the TRUCE and SUSHI datasets, which pairtime-series signals with natural language descriptions, we demonstrate thatCLaSP achieves high accuracy in retrieving a variety of time series patternsbased on natural language queries.