ForPKG: A Framework for Constructing Forestry Policy Knowledge Graph and Application Analysis

TL;DR

This paper tackles the challenge of building a policy knowledge graph in forestry by proposing a complete framework (ForPKG) that combines a fine-grained ontology with an unsupervised, open-source LLM-based information extraction pipeline. It defines 10 entity types and 15 relation types enriched with deontic-logic concepts, constructs both document- and content-level knowledge, and demonstrates improved extraction performance (e.g., 76.2% precision, 62.6% recall) over baselines. It also validates the KG's practical value in retrieval-augmented generation tasks using LLMs and showcases integration with general knowledge graphs to boost query accuracy. The forestry policy KG is released on Github to support forestry policy analysis, compliance, and intelligent systems, while offering a blueprint for policy KG construction in other domains.

Abstract

A policy knowledge graph can provide decision support for tasks such as project compliance, policy analysis, and intelligent question answering, and can also serve as an external knowledge base to assist the reasoning process of related large language models. Although there have been many related works on knowledge graphs, there is currently a lack of research on the construction methods of policy knowledge graphs. This paper, focusing on the forestry field, designs a complete policy knowledge graph construction framework, including: firstly, proposing a fine-grained forestry policy domain ontology; then, proposing an unsupervised policy information extraction method, and finally, constructing a complete forestry policy knowledge graph. The experimental results show that the proposed ontology has good expressiveness and extensibility, and the policy information extraction method proposed in this paper achieves better results than other unsupervised methods. Furthermore, by analyzing the application of the knowledge graph in the retrieval-augmented-generation task of the large language models, the practical application value of the knowledge graph in the era of large language models is confirmed. The knowledge graph resource will be released on an open-source platform and can serve as the basic knowledge base for forestry policy-related intelligent systems. It can also be used for academic research. In addition, this study can provide reference and guidance for the construction of policy knowledge graphs in other fields. Our data is provided on Github https://github.com/luozhongze/ForPKG.

Figures (6)

• Figure 1: The workflow for building a forestry policy knowledge graph using an open-source LLM.
• Figure 2: The improvement in response accuracy after merging the forestry policy knowledge graph with five general knowledge graphs.
• Figure 3: Accuracy on different entity types and relationship types.
• Figure 4: Training sample dependency analysis.
• Figure 5: Evaluation results of answer generation quality before and after integrating the LLaMa-Chinese model with the knowledge graph of forestry policies.