COMMIT at SemEval-2017 Task 5: Ontology-based Method for Sentiment Analysis of Financial Headlines

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 883–887, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics COMMIT at SemEval-2017 Task 5: Ontology-based Method for Sentiment Analysis of Financial Headlines Kim Schouten Flavius Frasincar Erasmus University Rotterdam P.O. Box 1738, NL-3000 DR Rotterdam, The Netherlands {schouten,frasincar}@ese.eur.nl f.m.g.dejong@eshcc.eur.nl Franciska de Jong Abstract This paper describes our submission to Task 5 of SemEval 2017, Fine-Grained Sentiment Analysis on Financial Mi- croblogs and News, where we limit our- selves to performing sentiment analysis on news headlines only (track 2). The ap- proach presented in this paper uses a Sup- port Vector Machine to do the required regression, and besides unigrams and a sentiment tool, we use various ontology- based features. To this end we created a domain ontology that models various con- cepts from the financial domain. This al- lows us to model the sentiment of actions depending on which entity they are affect- ing (e.g., decreasing debt is positive, but decreasing profit is negative). The pre- sented approach yielded a cosine distance of 0.6810 on the official test data, resulting in the 12th position. 1 Introduction Many companies in the financial sector are in the business of gathering and selling information, in- cluding news and sentiment analysis information, because of the profound influence this has on in- vestor behavior (Van de Kauter et al., 2015). The relation between news and movements in the fi- nancial market is intricate, with news influencing the market (Schuster, 2003) as well as the mar- ket being a source of news itself. Price fluctu- ations of financial instruments can be linked to supply and demand and thus to the desirability of that financial product, which changes when new facts related to this product are published. Sen- timent analysis in the context of financial news headlines aims to capture the change in desirabil- ity of a given product. Assigning a negative senti- ment score to a certain news headline for a given product then represents a decrease in desirabil- ity and thus a decrease in price for that product, while assigning a positive sentiment has the oppo- site meaning. In track 2 of Task 5 at SemEval 2017 (Cortis et al., 2017), each news headline contains one or more company names, and for a given company name, the sentiment, modeled as a real number between -1 and 1, needs to be determined. When multiple companies are mentioned, the same sen- tence can appear multiple times in the data, each time asking for the sentiment with respect to a dif- ferent company. In financial headlines, there are two reasons why the expressed sentiment can dif- fer for the various companies that are mentioned. The first is that expressed sentiment is often oppo- site for competitors, while the second is that news often reflects on the stock movements of the day mentioning both the biggest winners and biggest losers in the same headline. Besides directly mentioning the stock move- ments of a company, …

Content truncated to 500 words