Construction of Q&A methods based on knowledge graphs and large language models-improving the accuracy of landscape pest and disease Q&A

Abstract

With the development of urban landscaping, the problem of garden diseases and pests is becoming increasingly severe. Large language models have garnered significant attention for their ability to understand user intent and provide answers. The introduction of knowledge graphs has provided a high quality knowledge base for large language models. This study combines knowledge graphs (KGs), large language models (LLMs) and other technologies to design an intelligent question-answering (Q&A) model for garden pests and diseases. The main work carried out is as follows:

Build a knowledge graph for garden diseases and pests by collecting high-quality data through web crawling and literature analysis. Identify key entities and relationships to construct a conceptual pattern layer. Applying the ERNIE-BiLSTM-CRF model to extract knowledge from unstructured data. Through experiments, it is found that the accuracy, recall and F1 value of the knowledge extraction model proposed in this study are all more than 92%, superior to other models.

Propose a Q&A method that integrates the garden pest and disease KG with the ERNIE-Bot-turbo model. By vectorizing the knowledge and using similarity matching, the most relevant data is retrieved, combined with the question to form prompts, and input into the language model to generate natural language answers. Experiments comparing our method with ERNIE-Bot-turbo and ChatGLM-6B showed that our approach performs well on simple, moderate, and complex problems, avoiding misleading answers for irrelevant questions. It outperforms both models in accuracy, achieving a 90% accuracy rate for simple questions.