Text mining has emerged as an important suite of techniques in recent years and found its way into many applications. In the finance area, most recent text mining-based studies focus on the prediction of the stock market trend or the detection of company bankruptcy/fraud. Other important economic indicators of the companies, such as revenues, are seldom addressed. Yet these indicators could be quite important and reflect the financial status of the company’s cash flow and market share. In this paper, we adopt a lexicon-based approach that first builds several lexicons of different types, including sources, entities, aspects, sentiments, and past times. Twelve sentiment features are identified as predictors of revenue trend and a lexicon-based method for determining the sentiment of each feature is proposed. In addition, one more feature computed using ARIMA based on previous revenue data is incorporated. Our experimental results using news articles of the seven Taiwan-based, major PC manufacturing companies demonstrate that both financial news articles and previous revenue data are important for accurately predicting revenue trend. The prediction model constructed using the proposed approach is able to predict revenue trend with accuracy of more than 80%.
Huang, Hsin-Ching; Hwang, San-Yih; Chang, Shanlin; and Kang, Yihuang, "Forecasting Company Revenue Trend Using Financial News" (2017). PACIS 2017 Proceedings. 193.