Sentiment Analysis for Financial Applications: Combining Machine Learning, Computational Linguistics, and Statistical Methods for Predicting Stock Price Behavior

摘要: In this thesis we use sentiment analysis, a classification task within the field of artificial intelligence, for financial applications. Hereunder, combine machine learning, computational linguistics, and statistical methods anticipating stock price behavior ten shares listed on Oslo Stock Exchange (OSE). These predictions have been made basis classifications firm-specific news articles, output by our specially constructed engine, an aggregated market-wide index. The motivation approach comes from being most felicitous source information; in effect widely-read filtering aggregating funnel sentiments. Furthermore, OSE has selected, firstly, its faculty inefficient, compared to peer marketplaces, and, secondly, inherent barriers processing Norwegian language associated with exchange, having meagre linguistic resources. If able surmount these exploit predictive value sentiments, one could potentially attain competitive advantage trading market. constructing named found contextual features be paramount precision addition developed optimized parsimonious lexica construction. Despite lack resources, achieve state-of-the-art using manual annotation. engine make statistically significant return, volume, order size. Positive predominantly, lead increases volume while negative articles predict opposite effect. same is general proclivity For only impact future behavior, ceteris paribus, depreciating subsequent prices. interaction between also significant. Although sign latter seems firm-idiosyncratic, analysis reveal that illiquid stocks exhibit stronger reactions than liquid stocks.