Chow-Liu Ordering for Long-Context Reasoning in Chain-of-Agents

TL;DR

This paper introduces a method using Chow-Liu trees to optimize chunk ordering in multi-agent reasoning, significantly improving answer relevance and accuracy in long-context tasks by reducing information loss.

Contribution

It proposes a novel chunk ordering strategy based on Chow-Liu trees to enhance long-context reasoning in multi-agent frameworks, addressing information loss issues.

Findings

Chow-Liu tree-based ordering outperforms default and semantic score-based orderings.

Breadth-first traversal of the Chow-Liu tree reduces information loss.

Improved accuracy across three long-context benchmarks.

Abstract

Sequential multi-agent reasoning frameworks such as Chain-of-Agents (CoA) handle long-context queries by decomposing inputs into chunks and processing them sequentially using LLM-based worker agents that read from and update a bounded shared memory. From a probabilistic perspective, CoA aims to approximate the conditional distribution corresponding to a model capable of jointly reasoning over the entire long context. CoA achieves this through a latent-state factorization in which only bounded summaries of previously processed evidence are passed between agents. The resulting bounded-memory approximation introduces a lossy information bottleneck, making the final evidence state inherently dependent on the order in which chunks are processed. In this work, we study the problem of chunk ordering for long-context reasoning. We use the well-known Chow-Liu trees to learn a dependency structure that prioritizes strongly related chunks. Empirically, we show that a breadth-first traversal of the resulting tree yields chunk orderings that reduce information loss across agents and consistently outperform both default document-chunk ordering and semantic score-based ordering in answer relevance and exact-match accuracy across three long-context benchmarks.