In order to better excavate investor sentiment, this paper solves the problems of the existing sentiment dictionary construction methods in the stock market text sentiment analysis process, such as low degree of automation, insufficient industry specificity and insufficient accuracy, etc.. Based on the construction of the basic sentiment dictionary, Word2vec automatically add high-frequency sentiment words for polarity judgment and assignment. Construction of the sentiment dictionary is changed into an optimization problem. An improved simulated annealing algorithm is adopted to optimize the word scores of the sentiment dictionary and to boost the performance of the stock market sentiment dictionary. The experimental results show that the stock market sentiment dictionary constructed by this method can effectively identify stock market text sentiment, improve the accuracy of sentiment analysis, and be better used in investor sentiment related research.