The evaluation of news articles' polarity from the stock price

Abstract

Many stuides have been attempting to uncover the relationship between asset price fluctuations and textual information. However, in many cases, subjective evaluations of researchers have significant impacts on results. In this paper, we propose an objective evaluation method of news article using stock prices. Firstly, we give class labels (negative-positive) to every news articles by using event study analysis. Secondly, we compute vector representations of news aritcles by morphological analysis and tf-idf. Finally, we make Support Vector Machine classifier and give class labels (negative-positive) to other news articles. This method enable objective analyses of textual data in financial contexts.