Abstract: With the development of Internet technology,using data mining techniques can not only construct and measure high-frequency emotional indicators of investors,but also help in-depth study of the intrinsic linkage between investor sentiment and stock market operations. The paper selects a total of 2 198 trading days in 2009-2018,using reptile software to excavate the network data reviewed by the Stock Exchange Index,using natural language processing techniques for text analysis,and constructing investor sentiment diurnal indices,uses DCC-GARCH model to study the time-varying correlation between investor sentiment,market excess returns and market liquidity. The empirical results show that there is a time-varying correlation between investor sentiment and market excess returns and liquidity,and the dynamic condition correlation coefficient is positive overall. This correlation has long-term memory,and it can be affected by the macroeconomic factors; Compared to the bull market,the operation of the stock market in a bear market environment is more sensitive to changes in investor sentiment. Compared to the monthly data,the higher-frequency daily indicator can capture instant and accurate information when describing sentiment and stock market relevance. These conclusions have certain significance for deep understanding of the linkage mechanism between investor sentiment and stock returns and liquidity.