LaSTR: Language-Driven Time-Series Segment Retrieval

2026-02-28

AI Summary

Existing time series retrieval methods struggle to accurately localize fine-grained segments based on natural language queries, as they rely on expert-designed global similarity metrics. This work proposes LaSTR, the first framework for language-driven, fine-grained time series segment retrieval. LaSTR leverages the TV2 segmentation algorithm to extract local segments and employs GPT to generate high-quality textual descriptions, thereby constructing a large-scale paired dataset. Built upon a Conformer architecture and trained with contrastive learning, LaSTR aligns time series segments and natural language in a shared embedding space to achieve cross-modal semantic alignment. Experiments demonstrate that LaSTR significantly outperforms random and CLIP-based baselines across various candidate pool sizes, substantially improving retrieval ranking quality and semantic consistency.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Computer Vision: Image and Video Retrieval
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Effectively searching time-series data is essential for system analysis, but existing methods often require expert-designed similarity criteria or rely on global, series-level descriptions. We study language-driven segment retrieval: given a natural language query, the goal is to retrieve relevant local segments from large time-series repositories. We build large-scale segment--caption training data by applying TV2-based segmentation to LOTSA windows and generating segment descriptions with GPT-5.2, and then train a Conformer-based contrastive retriever in a shared text--time-series embedding space. On a held-out test split, we evaluate single-positive retrieval together with caption-side consistency (SBERT and VLM-as-a-judge) under multiple candidate pool sizes. Across all settings, LaSTR outperforms random and CLIP baselines, yielding improved ranking quality and stronger semantic agreement between retrieved segments and query intent.

Problem

• time-series retrieval
• language-driven retrieval
• segment retrieval
• natural language query
• time-series analysis

Innovation

• language-driven retrieval
• time-series segmentation
• contrastive learning
• Conformer
• multimodal embedding