Financial News Sentiment Analysis Method Based on WMSA-Bi-LSTM

In order to analyze and judge the sentiment of financial market more accurately, a financial news sentiment analysis model WMSA-Bi-LSTM is proposed. First of all, the financial market sentiment lexicon is established to assist Word2Vec to convert the input text data into word vector representation in line with the financial direction. Then, Bi-LSTM is used to extract the context information and important features of financial news text data. Finally, Multi-head Self-Attention mechanism is added to enhance the ability of the model to extract key features. In the SAFN dataset and the SACFN dataset, the Acc of WMSA-Bi-LSTM is 82.19%, 74.93%, and F1 is 82.22%, 75.05%, respectively, superior to RNN, LSTM and GRU series models.