Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mixture of Knowledge Minigraph Agents for Literature Review Generation

Zhi Zhang, Yan Liu, Sheng-hua Zhong, Gong Chen, Yu Yang, Jiannong Cao

TL;DR

This paper tackles the time-intensive task of literature review by introducing collaborative knowledge minigraph agents (CKMAs) that explicitly model cross-paper relations to guide abstractive summaries. The Knowledge Minigraph Construction Agent (KMCA) builds concise, topic-focused knowledge minigraphs from reference abstracts in iterative chunks, while the Multiple Path Summarization Agent (MPSA) generates and merges multiple viewpoint summaries through a mixture-of-experts framework and a self-consistency-based router. Across three MSDS benchmarks, CKMAs achieve state-of-the-art ROUGE scores, with ablations showing substantial gains from both KMCA and MPSA components, particularly from iterative minigraph construction and path-aware routing. The work demonstrates that structured, graph-informed prompting of LLMs can produce more faithful, concise, and comprehensive literature reviews, suggesting strong practical potential for AI-assisted scientific synthesis.

Abstract

Literature reviews play a crucial role in scientific research for understanding the current state of research, identifying gaps, and guiding future studies on specific topics. However, the process of conducting a comprehensive literature review is yet time-consuming. This paper proposes a novel framework, collaborative knowledge minigraph agents (CKMAs), to automate scholarly literature reviews. A novel prompt-based algorithm, the knowledge minigraph construction agent (KMCA), is designed to identify relations between concepts from academic literature and automatically constructs knowledge minigraphs. By leveraging the capabilities of large language models on constructed knowledge minigraphs, the multiple path summarization agent (MPSA) efficiently organizes concepts and relations from different viewpoints to generate literature review paragraphs. We evaluate CKMAs on three benchmark datasets. Experimental results show the effectiveness of the proposed method, further revealing promising applications of LLMs in scientific research.

Mixture of Knowledge Minigraph Agents for Literature Review Generation

TL;DR

This paper tackles the time-intensive task of literature review by introducing collaborative knowledge minigraph agents (CKMAs) that explicitly model cross-paper relations to guide abstractive summaries. The Knowledge Minigraph Construction Agent (KMCA) builds concise, topic-focused knowledge minigraphs from reference abstracts in iterative chunks, while the Multiple Path Summarization Agent (MPSA) generates and merges multiple viewpoint summaries through a mixture-of-experts framework and a self-consistency-based router. Across three MSDS benchmarks, CKMAs achieve state-of-the-art ROUGE scores, with ablations showing substantial gains from both KMCA and MPSA components, particularly from iterative minigraph construction and path-aware routing. The work demonstrates that structured, graph-informed prompting of LLMs can produce more faithful, concise, and comprehensive literature reviews, suggesting strong practical potential for AI-assisted scientific synthesis.

Abstract

Literature reviews play a crucial role in scientific research for understanding the current state of research, identifying gaps, and guiding future studies on specific topics. However, the process of conducting a comprehensive literature review is yet time-consuming. This paper proposes a novel framework, collaborative knowledge minigraph agents (CKMAs), to automate scholarly literature reviews. A novel prompt-based algorithm, the knowledge minigraph construction agent (KMCA), is designed to identify relations between concepts from academic literature and automatically constructs knowledge minigraphs. By leveraging the capabilities of large language models on constructed knowledge minigraphs, the multiple path summarization agent (MPSA) efficiently organizes concepts and relations from different viewpoints to generate literature review paragraphs. We evaluate CKMAs on three benchmark datasets. Experimental results show the effectiveness of the proposed method, further revealing promising applications of LLMs in scientific research.

Paper Structure

This paper contains 13 sections, 4 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Graphs in MSDS Tasks
  4. Pre-trained Language Models in MSDS Tasks
  5. Method
  6. Knowledge Minigraph Construction Agent
  7. Multiple Path Summarization Agent
  8. Experiments
  9. Comparison Experiments
  10. Ablation Studies
  11. Case Studies
  12. Disscussion
  13. Conclusions and Future Work

Figures (5)

  • Figure 1: Illustration of relations between concepts in literature review.
  • Figure 2: The overall architecture of the proposed collaborative knowledge minigraph agents (CKMAs).
  • Figure 3: ROUGE score comparison of the proposed CKMAs with its backbone models (GPT-3.5-turbo) group by reference paper number on the Multi-Xscience dataset.
  • Figure 4: Case study of knowledge graphs queried from Wikidata (top) and knowledge minigraph constructed by CKMAs (bottom) for the topic "image denoising".
  • Figure 5: Performance comparison of CKMAs with its backbone model (GPT-3.5-turbo) along faithfulness, completeness, and conciseness dimensions. A higher score indicates better performance.