Sentiment Analysis on Weibo Platform for Stock Prediction

Abstract

Recent years has witnessed a growing trend on the stock prediction through sentiment analysis of social media. This paper proposes a stock-predicting method based on the sentiment analysis of Weibo, utilizing sentiment orientation exhibited in texts posted on Weibo and analysis on the historical data of stock market to forecast the price movements in stock market. To achieve the goal, three steps are strictly followed, corresponding to three parts of the paper: data pre-processing (DP), sentiment analysis (SA) and stock prediction (SP). In the data pre-processing phase, the construction of a financial dictionary is elaborated, and then, an improved LDA model is shown to classify Weibo text in accordance with Industrial Classification Benchmark (ICB) and to match them with corresponding topics. Then, the sentiment analysis part demonstrates that how the accuracy of sentiment analysis is enhanced through the newly designed method by incorporating the interaction between words to the rule set to quantify sentiment value. In the last part, the stock prediction is accomplished by using Multivariate SVM incorporating Weibo sentiment. The time range of collected Weibo data is 12 months.