Stock market commentary data on social media can reflect investors’ state of psychology about the stock market, and investor psychoanalysis is critical to predicting the direction of the stock market. Current text sentiment analysis only mines information from social media texts and lacks the combination of real stock market data, resulting in inadequate sentiment analysis for investors. The research on text sentiment analysis tends to focus on emotions and lacks the mining of psychology, which affects the accurate grasp of investor psychology. To this end, we construct a social media dataset containing stock market data and propose an investor psychological analysis method based on deep learning and real stock market data for more accurate investor psychological judgment. We first collect 220,000 comment data based on the Weibo A-share topic, then deduplicate, clean, and psychologically label the data. Subsequently, we collect stock market data over the comment period and construct the Psy-Stock dataset. Finally, we propose a method to analyze the psychological state of investors by fusing stock market data with textual features extracted from a pre-trained model, named BERT-Stock United Learning. After extensive experimental validation, our proposed methodology shows significant advantages over multiple baseline models in investor psychology analysis.