Black-Litterman Portfolio Study Based On BERT-BiLSTM Text Sentiment Mining

In recent years,with the development of network technology and internet finance,the dissemination of information in the financial market has become more convenient and efficient.Against this backdrop,investors’ demand for information has also been increasing.To provide more effective investment strategies,more and more scholars have begun to focus on the important influence of online investor sentiment on the financial market.A recent trend is to use text mining and deep learning techniques to analyze the sentiment of online financial comments and news articles to better understand market sentiment and investor sentiment in general,helping investors make optimal trading decisions.In this article,we start from the perspective of data mining and use Python programming to obtain stockholders’ post information on online platforms,constructing a BERT-Bi LSTM architecture to characterize the sentiment features of financial text statements.BERT pre-training models are used as underlying feature extractors to generate semantic vectors based on the dynamic features of text data context.These semantic vectors are then input into the Bi LSTM neural network to obtain bidirectional semantic dependency relationships.Then,the PCA principal component analysis method is used to construct a composite investor sentiment index.In addition,the dynamic relationship between investor sentiment and stock market returns is studied through correlation analysis and Granger causality tests.Finally,an ARIMAX model is used to adjust the expected returns of the investor sentiment index,generating investors’ subjective return vectors,which are applied to the Black-Litterman portfolio model for market asset allocation research.The results show that compared to traditional machine learning models such as LSTM and SVM,the sentiment classification model based on the BERT-Bi LSTM architecture has higher prediction accuracy and can more effectively identify the emotional color conveyed in financial text.In addition,the investor sentiment index constructed in this study is basically similar to the trend of stock returns,and stock returns will fluctuate with investor sentiment shocks and tend to stabilize in the short term.Finally,by reflecting the viewpoint returns with the investor sentiment index and applying it to the Black-Litterman investment portfolio model,the asset allocation effect is significantly improved,and the daily return rate and Sharpe ratio are both significantly increased.The innovation of this study lies in three aspects: first,the combination of the BERT model and the deep learning Bi LSTM model is used to perform sentiment analysis on unstructured financial text data,effectively improving the accuracy of financial text sentiment classification.Second,a multidisciplinary fusion method is used to construct an investor sentiment index,which more intuitively and comprehensively displays investors’ psychology and behavior.Third,in terms of investment decision-making,using the investor sentiment index to adjust the viewpoint returns of investors is more realistic in reflecting investors’ psychological expectations than models that rely solely on historical data for prediction.This method extends the theoretical framework of the Black-Litterman(BL)model and improves asset allocation effectiveness.Therefore,applying investor sentiment analysis to investment strategy research can improve the accuracy and efficiency of investment decisions,reduce investment risks,and increase portfolio performance and risk management capabilities,which has important research value.