Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Graph-Augmented Reasoning with Large Language Models for Tobacco Pest and Disease Management

Siyu Li, Chenwei Song, Qi Zhou, Wan Zhou, Xinyi Liu

TL;DR

This work tackles the challenge of domain-grounded reasoning for tobacco pest and disease management by integrating a domain knowledge graph with a large language model through a GraphRAG-style pipeline. It constructs a tobacco-specific graph, learns entity and relation representations with TransE and refines them via a Graph Convolutional Network, and fuses query-focused graph evidence into a Transformer-based generator to produce evidence-aware recommendations. Experimental results show consistent improvements over text-only baselines and other RAG variants, with the largest gains on multi-hop and comparative queries that require chaining symptom–disease–treatment relations. The approach demonstrates that explicitly modeling relational structure can enhances reliability and practical applicability of agricultural decision-support systems.

Abstract

This paper proposes a graph-augmented reasoning framework for tobacco pest and disease management that integrates structured domain knowledge into large language models. Building on GraphRAG, we construct a domain-specific knowledge graph and retrieve query-relevant subgraphs to provide relational evidence during answer generation. The framework adopts ChatGLM as the Transformer backbone with LoRA-based parameter-efficient fine-tuning, and employs a graph neural network to learn node representations that capture symptom-disease-treatment dependencies. By explicitly modeling diseases, symptoms, pesticides, and control measures as linked entities, the system supports evidence-aware retrieval beyond surface-level text similarity. Retrieved graph evidence is incorporated into the LLM input to guide generation toward domain-consistent recommendations and to mitigate hallucinated or inappropriate treatments. Experimental results show consistent improvements over text-only baselines, with the largest gains observed on multi-hop and comparative reasoning questions that require chaining multiple relations.

Graph-Augmented Reasoning with Large Language Models for Tobacco Pest and Disease Management

TL;DR

This work tackles the challenge of domain-grounded reasoning for tobacco pest and disease management by integrating a domain knowledge graph with a large language model through a GraphRAG-style pipeline. It constructs a tobacco-specific graph, learns entity and relation representations with TransE and refines them via a Graph Convolutional Network, and fuses query-focused graph evidence into a Transformer-based generator to produce evidence-aware recommendations. Experimental results show consistent improvements over text-only baselines and other RAG variants, with the largest gains on multi-hop and comparative queries that require chaining symptom–disease–treatment relations. The approach demonstrates that explicitly modeling relational structure can enhances reliability and practical applicability of agricultural decision-support systems.

Abstract

This paper proposes a graph-augmented reasoning framework for tobacco pest and disease management that integrates structured domain knowledge into large language models. Building on GraphRAG, we construct a domain-specific knowledge graph and retrieve query-relevant subgraphs to provide relational evidence during answer generation. The framework adopts ChatGLM as the Transformer backbone with LoRA-based parameter-efficient fine-tuning, and employs a graph neural network to learn node representations that capture symptom-disease-treatment dependencies. By explicitly modeling diseases, symptoms, pesticides, and control measures as linked entities, the system supports evidence-aware retrieval beyond surface-level text similarity. Retrieved graph evidence is incorporated into the LLM input to guide generation toward domain-consistent recommendations and to mitigate hallucinated or inappropriate treatments. Experimental results show consistent improvements over text-only baselines, with the largest gains observed on multi-hop and comparative reasoning questions that require chaining multiple relations.
Paper Structure (18 sections, 8 equations, 1 table)

This paper contains 18 sections, 8 equations, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Knowledge Graphs for Agricultural Pest and Disease Control.
  4. Large Language Models and Graph-Enhanced Reasoning.
  5. Knowledge Graphs and LLMs in Agricultural Domains.
  6. Methodology
  7. Knowledge Graph Construction
  8. Graph Embedding with TransE
  9. Node Representation Learning with GCN
  10. Integration with Large Language Models
  11. GraphRAG-Based Fusion
  12. Summary
  13. Experiments
  14. Experimental Setup
  15. Baselines and Evaluation Metrics
  16. ...and 3 more sections