Financial Fake News Detection with Multi fact CNN-LSTM Model

Abstract

Nowadays, financial news is an indispensable source for investors to conduct research and investment decisions. At the same time, there are many fake financial news flooded into people's daily life. This kind of information may affect public opinion and provide opportunities for some criminals to manipulate the financial market. However, due to the lack of available comparative information, the model based on linguistic features is much less effective in the real world. We believe that multi-source fact comparison and inspection should be integrated into the false news detection model to detect fake news. As the crystallization of collective wisdom, user comments can be of great benefit to this task. News sources are also crucial for detecting. Besides, existing models often ignore one point that financial fake news usually talks about the relevant market, so the market data should be token into consideration. Our proposed multi fact CNNLSTM model integrates all these dimensions mentioned above and performs well. Specially, we use attention mechanism to extract the information from the comments and make a list of authoritative websites to identify the source of news. As for the market dimension, according to the financial products mentioned in the news, we get market price and check whether the statements in the article are correct. Finally, we assign a weight to each dimension and let the model learns by itself.