山西大学学报（自然科学版）2026，Vol.49Issue(2)：189-198,10.DOI:10.13451/j.sxu.ns.2025101
基于多角度分析的大语言模型虚假信息检测
Large Language Models for Misinformation Detection Based on Multi-angle Analysis
摘要
Abstract
The spread of misinformation on social media is continual,however,existing methods based on deep learning and graph neural networks still face limitations in adapting to complex environments and providing interpretability.In addition,although large language models(LLMs)possess powerful language understanding capabilities,their potential in handling complex textual cues and propagation patterns has not been fully explored.This paper proposes a multi-angle feature extraction framework based on large lan-guage models,aiming to fully leverage the deep semantic mining and interpretability advantages of LLMs.Specifically,this frame-work guides LLMs to analyze social media content from ten perspectives,such as writing style,factual consistency,and netizen's sentiment,and outputs detailed explanatory responses by designing a set of multi-angle prompting instructions.The model then out-puts detailed responses along with explanatory rationales for each question.Subsequently,the framework utilizes the feature extrac-tion function of LLM to obtain the semantic representation of the session chain and the semantic representation of the interpreted text of the raw data,and further processes them with classification training using a multilayer perceptron,thus realizing low-cost and high-efficiency misinformation detection.Experimental results demonstrate that the proposed method in this paper significantly sur-passes two categories of baseline methods,Graph Neural Networks(GNNs)and Large Language Models(LLMs),in key metrics such as F1-score and accuracy.On the three public datasets of Weibo,Twitter15,and Twitter16,the proposed method in this paper improves the F1-score by 1.4%,3.8%and 3.5%,respectively,compared to the strongest baseline.Furthermore,the method not only maintains high performance but also substantially reduces training costs while enhancing the transparency and interpretability of model decision-making.关键词
社交媒体分析/谣言检测/人工智能/提示工程/可解释性Key words
social media analysis/rumor detection/artificial intelligence/prompt engineering/interpretability分类
信息技术与安全科学引用本文复制引用
贾彩燕,杨子琦,赵一,白祥意..基于多角度分析的大语言模型虚假信息检测[J].山西大学学报（自然科学版）,2026,49(2):189-198,10.基金项目
国家自然科学基金(62576026) （62576026）
中央高校基础科研业务项目(2024XKRC024) （2024XKRC024）
