[论文解读] SpatCode: Rotary-based Unified Encoding Framework for Efficient Spatiotemporal Vector Retrieval

Bingde Hu, Enhao Pan+4 more|arXiv (Cornell University)|Jan 14, 2026

Data Management and Algorithms被引用 0

一句话总结

SpatCode 将时间、空间和语义线索统一为一个旋转编码向量空间，实现通过圆形滑动窗口更新器和加权多模态排序的高效时空检索。它在多个真实数据集上比基线具有更高的召回率和更低的延迟。

ABSTRACT

Spatiotemporal vector retrieval has emerged as a critical paradigm in modern information retrieval, enabling efficient access to massive, heterogeneous data that evolve over both time and space. However, existing spatiotemporal retrieval methods are often extensions of conventional vector search systems that rely on external filters or specialized indices to incorporate temporal and spatial constraints, leading to inefficiency, architectural complexity, and limited flexibility in handling heterogeneous modalities. To overcome these challenges, we present a unified spatiotemporal vector retrieval framework that integrates temporal, spatial, and semantic cues within a coherent similarity space while maintaining scalability and adaptability to continuous data streams. Specifically, we propose (1) a Rotary-based Unified Encoding Method that embeds time and location into rotational position vectors for consistent spatiotemporal representation; (2) a Circular Incremental Update Mechanism that supports efficient sliding-window updates without global re-encoding or index reconstruction; and (3) a Weighted Interest-based Retrieval Algorithm that adaptively balances modality weights for context-aware and personalized retrieval. Extensive experiments across multiple real-world datasets demonstrate that our framework substantially outperforms state-of-the-art baselines in both retrieval accuracy and efficiency, while maintaining robustness under dynamic data evolution. These results highlight the effectiveness and practicality of the proposed approach for scalable spatiotemporal information retrieval in intelligent systems.

研究动机与目标

• Motivate unified spatiotemporal vector retrieval that avoids post-filtering and multi-index pipelines.
• Introduce a Rotary-based Unified Encoding that embeds time and location into rotational vectors.
• Propose a Circular Incremental Update Mechanism for efficient sliding-window data maintenance.
• Develop a Weighted Interest-based Retrieval Algorithm for adaptive, personalized multimodal ranking.
• Demonstrate superior retrieval accuracy and efficiency over state-of-the-art baselines on real datasets.

提出的方法

• Encode time and geography as rotational/colored vectors to form a unified embedding space using E_t(t) and E_g(λ,φ).
• Maintain a circular sliding window with a circular buffer to enable fast incremental updates without global re-encoding.
• Concatenate and scale modality embeddings into a single query vector to support weighted cosine similarity in a single ANN search.
• Apply a weighted interest-based retrieval to adapt modality importance per query without changing stored embeddings.

实验结果

研究问题

• RQ1 Can time and location be embedded into a unified vector space to enable direct spatiotemporal similarity with semantic features?
• RQ2 How can we efficiently maintain and update a spatiotemporal index in streaming data without full re-encoding?
• RQ3 Does a weighted multimodal retrieval approach improve accuracy for heterogeneous, time-evolving data?
• RQ4 How does SpatCode compare to existing filter-based and multi-index approaches in latency and recall on real datasets?

主要发现

Method	Shopping Insert	Shopping Query	Craigslist Insert	Craigslist Query	Netflix Insert	Netflix Query	Bridge Insert	Bridge Query	VeRi Insert	VeRi Query
ThalDB	0.0501	$>$ 1000	0.520	$>$ 1000	0.0343	$>$ 1000	0.0022	$>$ 1000	0.0415	$>$ 1000
Filtered	3.88	143	3.78	11.5	3.95	61.7	3.80	17.9	3.77	32.54
Hybrid	3.82	8.07	3.90	5.70	3.97	5.89	3.79	6.95	3.70	5.56
SpatCode	3.69	5.24	3.62	4.03	3.84	4.21	3.63	3.81	3.60	3.87

• SpatCode achieves higher recall than baselines across Shopping, Craigslist, Netflix, Bridge, and VeRi datasets at k = 10, 50, 100.
• Single-traversal ANN search with SpatCode attains lower query latency than multi-branch or filtered approaches.
• Circular Incremental Update reduces latency spikes and maintains stable performance over thirteen months of streaming data.
• Weighted Interest-based Retrieval enables effective per-modality tuning, achieving recall close to 1.0 at several settings for some datasets.
• Across experiments, SpatCode consistently outperforms ThalDB, Filtered, and Hybrid baselines in recall and efficiency.