On Fake News Detection with LLM Enhanced Semantics Mining

Anonymous

Abstract: Large language models (LLMs) have emerged as valuable tools for enhancing textual features in various text-related tasks. In this paper, we assess the effectiveness of news embeddings from ChatGPT for detecting fake news and showcase that despite their initial performance slightly surpassing the pre-trained BERT model, they still lag behind the state-of-the-arts. This shortfall is attributed to the reliance on tokenized training text, which misses the complex narratives and subtleties that are crucial for identifying fake news. To capture these nuances, we probe the high-level semantic relations among the news pieces, real entities, and topics, which are modeled as a heterogeneous graph with nodes denoting different items and the relations are represented as edges. We then propose a Generalized Page-Rank model and a consistent learning criteria for mining the local and global semantics centered on each news piece through the adaptive propagation of features across the graph. Our model shows new state-of-the-art performance on five benchmark datasets and the effectiveness of the key ingredients is supported by extensive analysis. Our code is available at \url{https://github.com/LEG4FD/LEG4FD}.

Paper Type: long

Research Area: NLP Applications

Contribution Types: NLP engineering experiment, Data analysis

Languages Studied: English