Abstract: There are irrational phenomena related to investor sentiment in the stock market， and the quantitative study of investor sentiment and stock market is important for discovering stock market patterns and aiding investment forecasting. Based on the investor statements in forums， a CNN-TLDA hybrid model is innovatively built to quantify investor sentiment from multiple perspectives， and explore the interaction between investor sentiment and the stock market from both positive and topic. The roles of investor sentiment in forecasting are investigated based on LSTM (long short-term memory) network. It is shown that Chinese stock market investors are generally pessimistic， and both investor optimism and topics of interest are highly correlated with the stock market. The multi-perspective sentiment analysis reduces the prediction error to 41%. The results of the study can assist investors in their investment decisions and also provide a scientific reference for the analysis and utilization of individual investors′ sentiment in the stock market.

Key words: investor sentiment; CNN(convolutional neural network); LDA(latent Dirichlet allocation) model; LSTM(long short-term memory) network; stock market prediction