A Versioned Unified Graph Index for Dynamic Timestamp-Aware Nearest Neighbor Search

ICLR 2026 Conference Submission13400 Authors

Keywords: approximate nearest neighbor search, recommender systems, time series

TL;DR: We introduce TiGER, a unified graph-based method for fast and accurate time-filtered ANN search on dynamic datasets.

Abstract: We present TiGER (Time-Integrated Graph for Efficient Retrieval), a novel approach for performing fast time-aware approximate nearest neighbor searches on dynamic vector datasets with flexibility over any possible time range. Our proposed algorithm constructs and maintains a unified graph for all vectors by utilising an index structure based on integrated, versioned connectivity, enabling direct querying of arbitrary time intervals on the unified graph without traversing invalid vectors. This forgoes the need for post-search filtering or merging, or to construct and/or save separate graphs for each possible composite range. Empirical evaluations show that our method attains up to a 5x improvement in queries per second (QPS) without compromising accuracy over state-of-the-art baselines based on filtering or per-time-segment sub-graphs. We believe that this method will enable efficient temporal analysis across evolving datasets in real-time recommendation systems, log analysis, and any scenario that requires fast similarity search over dynamic, time-segmented data.

Primary Area: optimization

Submission Number: 13400