Query-Conditioned Graph Retrieval for Contextualized LLM Reasoning in Personalized Wearable Data
The article discusses a new framework called Wearable As Graph (WAG) designed for analyzing personalized wearable data using large language models (LLMs). WAG addresses the challenge of context selection by organizing wearable metrics into a personalized knowledge graph and retrieving relevant subgraphs for improved reasoning. Evaluation results indicate that WAG significantly outperforms existing methods in terms of effectiveness for LLM-driven analysis of wearable data.
- ▪WAG organizes wearable metrics and user-specific signals into a personalized knowledge graph.
- ▪The framework retrieves a query-conditioned subgraph to enhance reasoning over wearable data.
- ▪WAG achieved a 70% win rate over baseline methods in evaluations with real-world datasets.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Information Retrieval arXiv:2605.18763 (cs) [Submitted on 10 Apr 2026] Title:Query-Conditioned Graph Retrieval for Contextualized LLM Reasoning in Personalized Wearable Data Authors:Zhenyu Lu, Mahyar Abbasian, Amir M. Rahmani View a PDF of the paper titled Query-Conditioned Graph Retrieval for Contextualized LLM Reasoning in Personalized Wearable Data, by Zhenyu Lu and 2 other authors View PDF HTML (experimental) Abstract:Large language models (LLMs) are increasingly applied to analyzing wearable sensing data, which are long-term, multimodal, and highly personalized. A key challenge is context selection: providing insufficient context limits reasoning, while including all available data leads to inefficiency and degraded generation quality.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
