Abstract: In the Internet age, more and more investors are beginning to express their investment opinions in online communities, especially in financial investment communities. The resulting massive financial text data has high research value. How to apply these financial text data has become the current research hotspots in the field of financial investment. This article explores how to convert investor posts in the Eastmoney Stock Forum into corresponding sentiment indicators, and form investor opinions based on this, and builds a portfolio model that considers financial text sentiment information under the framework of the Black-Litterman model. Specifically, we first use web crawlers to crawl the post text data of FTSE China’s A50 constituent stocks from the Eastmoney Stock Forum, and perform data preprocessing. Then, the sentiment indicators of the post text is extracted by using the dictionary method and the Naive Bayes method. Furthermore, three indicators of sentiment index, stock closing price and trading volume are taken as characteristic variables, and the random forest regression algorithm is used to predict the future return rate of stocks. Finally, the predicted future return rate is taken as the investor’s point of view, and is put into the framework of Black-Litterman model to construct a new portfolio model considering the emotional information of financial text. The backtest results show that the financial text sentiment mining portfolio model based on the Naive Bayes method has better performance.

Key words: portfolio, Black-Litterman model, investor sentiment, text mining, random forest regression