Sentiment Analysis from Stock Market News in Romanian using Chaos Game Representation

A recently proposed methodology for authorship attribution is adapted in the current work for sentiment analysis. Furthermore, it is applied here for a non-English language, i.e. for Romanian. The procedure works at the character level, hence it does not depend on the language, although it is designed only for the languages that use the Latin alphabet. The data set used is taken from financial market news and it contains paragraphs that refer to two particular companies. In order to establish the ground truth for the sentiment scores, the text is translated into English and the sentiment analysis tool VADER is further used. The aim of the methodology is to build a regression model that fits the initial paragraphs with text in Romanian to the scores established by VADER and the results are encouraging.