Reproduce “TRACE: State-Aware Query Processing over Temporal Evidence Graphs for Conversational Data” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Maolin Wang, Yu Wang, Zichun Liu, Baiyuan Qiu, Chenbin Zhang, Jiguang Shen, Haoran Yang, Hao Miao
The TRACE framework introduces a state-aware query processing approach utilizing a temporal evidence graph to manage evolving conversational data for long-running AI agents. It demonstrated improved accuracy for temporal and multi-hop reasoning questions on long conversation QA benchmarks, outperforming baselines by up to 19.9 percentage points, while significantly reducing token consumption.
Key insights
- Explicitly modeling state evolution within a temporal evidence graph is crucial for accurately answering queries over dynamic conversational data by distinguishing between valid and obsolete information.
- Combining structured graph evidence with query-shape-aware reasoning instructions for LLMs significantly enhances performance on complex reasoning tasks, particularly temporal and multi-hop questions.
- A hybrid approach that integrates both lexical retrieval and path-based graph traversal, alongside efficient candidate gating and budget allocation, enables scalable and efficient state-aware query processing over extensive conversational histories.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.