TwiSe at SemEval-2017 Task 4: Five-point Twitter Sentiment Classification and Quantification

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 755–759, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics TwiSe at SemEval-2017 Task 4: Five-point Twitter Sentiment Classification and Quantification Georgios Balikas University of Grenoble Alps, CNRS, Grenoble INP - LIG georgios.balikas@imag.fr Abstract The paper describes the participation of the team “TwiSE” in the SemEval-2017 challenge. Specifically, I participated at Task 4 entitled “Sentiment Analysis in Twitter” for which I implemented systems for five-point tweet classification (Sub- task C) and five-point tweet quantification (Subtask E) for English tweets. In the fea- ture extraction steps the systems rely on the vector space model, morpho-syntactic analysis of the tweets and several sen- timent lexicons. The classification step of Subtask C uses a Logistic Regression trained with the one-versus-rest approach. Another instance of Logistic Regression combined with the classify-and-count ap- proach is trained for the quantification task of Subtask E. In the official leaderboard the system is ranked 5/15 in Subtask C and 2/12 in Subtask E. 1 Introduction Microblogging platforms like Twitter have lately become ubiquitous, democratizing the way people publish and access information. This vast amount of information that reflects the opinions, news or comments of people creates several opportunities for opinion mining. Among other platforms, Twit- ter is particularly popular for research due to its scale, representativeness and ease of access to the data it provides. Furthermore, to facilitate the study of opinion mining, high quality resources and data challenges are organized. The Task 4 of the SemEval-2017 challenges, entitled “Sentiment Analysis in Twitter” is among them. The paper describes the participation of the team Twitter Sentiment (TwiSe) in two of the subtasks of Task 4 of SemEval-2017. Specifically, I participated in Subtasks C and E. Both of them assume that sentiment is distributed across a five- point scale ranging from VeryNegative to VeryPos- itive. Subtask C is a sentiment classification task, where given a tweet the aim is to assign one of the five classes. Subtask E is a quantification task, whose aim is given a set of tweets referring to a subject to estimate the prevalence of each of the five classes. The tasks are described in more detail at (Rosenthal et al., 2017). The rest of the paper is organized as follows: Section 2 describes the feature extraction steps performed in order to construct the representation of a tweet, which is the same for both subtasks C and E. Section 3 details the learning approaches used and Section 4 summarizes the achieved per- formance. Finally, Section 5 concludes with point- ers for future work. 2 Feature Extraction In this section I describe the details of the feature extraction process performed. My approach is heavily inspired by my previous participation in the “Twitter Sentiment Analysis” task of SemEval-2016, which is detailed at Balikas and Amini (2016). Importantly, the code for perform- ing the feature extraction steps described below is publicly available at https://github. com/balikasg/SemEval2016-Twitter_ Sentiment_Evaluation. There are three sets of features extracted: 1. …

Content truncated to 500 words