DUTH at SemEval-2017 Task 5: Sentiment Predictability in Financial Microblogging and News Articles

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 861–865, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics DUTH at SemEval-2017 Task 5: Sentiment Predictability in Financial Microblogging and News Articles Symeon Symeonidis John Kordonis Dimitrios Effrosynidis Avi Arampatzis Database & Information Retrieval research unit, Department of Electrical & Computer Engineering, Democritus University of Thrace, Xanthi 67100, Greece {ssymeoni,ioankord1,dimievfr,avi}@ee.duth.gr Abstract We present the system developed by the team DUTH for the participation in Semeval-2017 task 5 - Fine-Grained Sen- timent Analysis on Financial Microblogs and News, in subtasks A and B. Our ap- proach to determine the sentiment of Mi- croblog Messages and News Statements & Headlines is based on linguistic pre- processing, feature engineering, and su- pervised machine learning techniques. To train our model, we used Neural Network Regression, Linear Regression, Boosted Decision Tree Regression and Decision Forrest Regression classifiers to forecast sentiment scores. At the end, we present an error measure, so as to improve the per- formance about forecasting methods of the system. 1 Introduction Social media sentiment is an important indica- tor of public opinion. Determining sentiment can be valuable in a number of applications including brand awareness, product launches, and detecting political trends. Many microblogging platforms such as Twitter and StockTwits have become very popular and are employed by many traders and in- vestors. Recently, many studies (Pi˜neiro-Chousa et al., 2016; Van de Kauter et al., 2015; Kordo- nis et al., 2016) used sentiment from social media and financial news articles trying to analyze mar- ket movements. This paper describes our submissions to Se- mEval 2017 task 5 (Cortis et al., 2017), which deals with sentiment analysis in microblog mes- sages for SubTask A, and sentences for news head- lines for SubTask B. In SubTask A, our model was ranked last because of a submission format error. We perform error measures in order to ob- tain a better understanding of the strengths of these particularly new tasks and to improve the perfor- mance about forecasting methods of our model (Armstrong and Collopy, 1992). For Subtask B, our team was ranked 24th from 29 teams. The rest of this paper is structured as follows: Section 2 provides our system’s description. Sec- tion 3 presents the experiments and some unoffi- cial results used in analyzing the system’s perfor- mance. Finally, conclusions and further directions for research are presented in Section 4. 2 System Description In this section we present the details of our senti- ment analysis system, feature extraction and some statistics about preprocessing. 2.1 Dataset The Task 5 organizers (Cortis et al., 2017) pro- vided a training and testing set for both subtasks. For subtask A, resources of Microblog messages were Stocktwits and Twitter, which have been an- notated for fine-grained sentiment 1. A collection of financially relevant news headlines which have been annotated for fine-grained sentiment, from sources such as Yahoo Finance, are given from Task Organizers for subtask B 2. Some statistics about the datasets are presented in …

Content truncated to 500 words