Abstract: To address the issue that existing fake news detection models overly relied on data collected from a single social platform for analysis, thereby neglecting differences in news content across various media platforms and lacking attention to features generated during the information dissemination process, a fake news detection model based on cross-social media dissemination features (FBCM) was proposed. First, relevant content and dissemination data related to the news to be detected were collected from different social platforms, and a cross-social media dataset was constructed. Then, the content data was analyzed using eight benchmark models to derive preliminary judgment results, while dissemination features were extracted from the dissemination data to obtain dissemination results. Finally, the two results were combined and integrated with post-heuristic strategies for error correction and adjustment to achieve the final judgment. Experimental results indicate that, when validated on cross-platform datasets, the proposed model achieves improvements of 8 percentage points in accuracy, 8 percentage points in precision, 15 percentage points in recall, and 11 percentage points in F1 score compared to the MPFN model. The issue of poor timeliness associated with existing fake news detection methods that rely on single-platform data is effectively addressed.

Key words: Natural Language Processing (NLP), fake news detection, cross- social media, online multi-source data, dissemination features

摘要： 针对现有虚假新闻检测模型依赖单一社交平台数据收集并进行分析，忽视不同媒体平台新闻内容差异，且对信息在传播过程中产生的特征缺乏关注的问题，提出一种基于跨社交媒体传播特征的虚假新闻检测模型(fake news detection model based on cross-social media dissemination features ,FBCM)。FBCM首先从不同社交平台收集待检测新闻的相关内容及传播数据，构建跨社交媒体数据集。其次将内容数据通过八个基准模型进行分析得出预判断结果，同时从传播数据中获取传播特征得出传播结果，最后结合两方面结果并融入后启发式思想进行纠错调整，得到最终判断。实验结果表明，在验证跨平台数据集时，本文提出的模型与 MPFN 模型相比，准确率提高了8个百分点，精确度提高了8个百分点，召回率提高了15个百分点，F1 分数提高了11个百分点，有效解决了现有假新闻检测方法依赖单一平台数据，时效性差的问题。

关键词: 自然语言处理, 虚假新闻检测, 跨社交媒体, 在线多源数据, 传播特征