Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2604.11610 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866913026144206848 |
|---|---|
| author | Yang, Yuqing Liu, Tengxiao Zhu, Wang Bill Shi, Taiwei Song, Linxin Jia, Robin |
| author_facet | Yang, Yuqing Liu, Tengxiao Zhu, Wang Bill Shi, Taiwei Song, Linxin Jia, Robin |
| contents | As LLM-based assistants become persistent and personalized, they must extract and retain useful information from past conversations as memory. However, the types of information worth remembering vary considerably across tasks. We formalize the \textit{heterogeneous memory extraction} task and introduce \textbf{BEHEMOTH}, a benchmark that repurposes 18 existing datasets spanning personalization, problem-solving, and agentic tasks, using a downstream utility-driven metric for systematic evaluation. Our empirical analysis confirms that no single static extraction prompt dominates across all task categories, and that existing self-evolving prompt optimization frameworks, originally designed for homogeneous distributions, degrade when training tasks are heterogeneous. To address this, we propose \textbf{CluE}, a cluster-based self-evolving strategy that groups training examples into clusters by extraction scenarios, analyzes each cluster independently, and synthesizes cross-cluster insights to update the extraction prompt. Experiments on BEHEMOTH show that CluE generalizes effectively across heterogeneous tasks ($+$9.04\% relative gain), consistently outperforming prior self-evolving frameworks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_11610 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Self-Evolving LLM Memory Extraction Across Heterogeneous Tasks Yang, Yuqing Liu, Tengxiao Zhu, Wang Bill Shi, Taiwei Song, Linxin Jia, Robin Computation and Language As LLM-based assistants become persistent and personalized, they must extract and retain useful information from past conversations as memory. However, the types of information worth remembering vary considerably across tasks. We formalize the \textit{heterogeneous memory extraction} task and introduce \textbf{BEHEMOTH}, a benchmark that repurposes 18 existing datasets spanning personalization, problem-solving, and agentic tasks, using a downstream utility-driven metric for systematic evaluation. Our empirical analysis confirms that no single static extraction prompt dominates across all task categories, and that existing self-evolving prompt optimization frameworks, originally designed for homogeneous distributions, degrade when training tasks are heterogeneous. To address this, we propose \textbf{CluE}, a cluster-based self-evolving strategy that groups training examples into clusters by extraction scenarios, analyzes each cluster independently, and synthesizes cross-cluster insights to update the extraction prompt. Experiments on BEHEMOTH show that CluE generalizes effectively across heterogeneous tasks ($+$9.04\% relative gain), consistently outperforming prior self-evolving frameworks. |
| title | Self-Evolving LLM Memory Extraction Across Heterogeneous Tasks |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2604.11610 |