TakeLab at SemEval-2017 Task 4: Recent Deaths and the Power of Nostalgia in Sentiment Analysis in Twitter

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 784–789, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics TakeLab at SemEval-2017 Task 4: Recent Deaths and the Power of Nostalgia in Sentiment Analysis in Twitter David Lozi´c, Doria Šari´c, Ivan Toki´c, Zoran Medi´c, Jan Šnajder Text Analysis and Knowledge Engineering Lab Faculty of Electrical Engineering and Computing, University of Zagreb Unska 3, 10000 Zagreb, Croatia {name.surname}@fer.hr Abstract This paper describes the system we sub- mitted to SemEval-2017 Task 4 (Sen- timent Analysis in Twitter), specifically subtasks A, B, and D. Our main focus was topic-based message polarity classi- fication on a two-point scale (subtask B). The system we submitted uses a Support Vector Machine classifier with rich set of features, ranging from standard to more creative, task-specific features, including a series of rating-based features as well as features that account for sentimental rem- iniscence of past topics and deceased fa- mous people. Our system ranked 14th out of 39 submissions in subtask A, 5th out of 24 submissions in subtask B, and 3rd out of 16 submissions in subtask D. 1 Introduction Sentiment analysis (Pang et al., 2002), a task of determining polarity of text towards some topic, recently gained a lot of interest, mostly due to its applicability in various fields, such as public rela- tions (Pang et al., 2008) and market analysis (He et al., 2013). Following the growing popularity of social networks and an increasing number of user comments that can be found there, sentiment anal- ysis of texts on social networks, such as tweets from Twitter, has been the focus of much research. However, determining the sentiment of a tweet is often not an easy task, since the length of the tweet is limited and language is mostly informal, including slang, abbreviations, and hashtags. Var- ious systems have been proposed for tackling this problem, ranging from simple unsupervised mod- els that use precompiled sentiment lexicons for evaluating polarity of tweets (O’Connor et al., 2010) to more complex supervised models that use textual feature representations in combination with machine learning algorithms such as Sup- port Vector Machines (SVM) (Khan et al., 2015; Barbosa and Feng, 2010) or deep neural networks (Dos Santos and Gatti, 2014; Tang et al., 2014). In this paper, we present our system for de- termining sentiment of tweets, which we submit- ted SemEval-2017 Task 4 (Rosenthal et al., 2017), more specifically to the English versions of sub- tasks A, B, and D. In subtask A, the goal was to predict the sentiment of a tweet as either positive, neutral, or negative. Subtask B consisted of pre- dicting the sentiment of a given tweet on a 2-point scale (positive or negative) given a topic. In sub- task D, the task was to determine the distribution of positive and negative tweets for each topic in a given set of tweets annotated with topics. The system we submitted uses an SVM clas- sifier with a linear kernel and a number …

Content truncated to 500 words