Food23:A Chinese Food Safety Dataset for Fake News Detection

In the field of fake news, the topic of food safety has gradually emerged as a focal point of concern. We have exhaustively scoured the relevant scholarly literature pertaining to the detection of fake news in the domain of food safety. Unfortunately, the extant research concerning the detection of food safety fake news remains focused on analyzing their social impact, rather than providing an effective detection method. In order to facilitate research on fake news detection in the domain of food safety, this study constructed the first Chinese text fake news dataset specific to the domain of food safety. We gathered data from currently active Chinese fact-checking websites and authoritative news sources as our primary reservoir of information. This dataset contains 2,334 news across 10 distinct subdomains. These specialized domains have effectively enriched the dataset's characteristics. According to this dataset, we design an efficient model for food safety fake news detection, known as the Food Safety Fake News Detection (FSFND). By extracting text features from various perspectives and considering the characteristics inherent to each subdomain, our model makes predictions regarding the veracity of information. We chose a selection of text classification models and multi-domain fake news detection model as comparative experiments. Experimental results indicate that our approach significantly outperforms conventional methods reliant on text and domain information for detecting food safety fake news. In addition, we conducted ablation experiments to demonstrate the effectiveness of each step in our model design.