Minions at SemEval-2016 Task 4: or how to build a sentiment analyzer using off-the-shelf resources?

Proceedings of SemEval-2016, pages 247–250, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics Minions at SemEval-2016 Task 4: or how to build a sentiment analyzer using off-the-shelf resources? Călin-Cristian Ciubotariu, Marius-Valentin Hrişca, Mihail Gliga, Diana Darabană, Diana Trandabăţ and Adrian Iftene University “Alexandru Ioan Cuza” of Iaşi, Romania {calin.ciubotariu, marius.hrisca, mihail.gliga, diana.darabana, diana.trandabat, adiftene}@info.uaic.ro Abstract Minions, a team formed of first year stu- dents in the Master of Computational Lin- guistics, started the participation at Semeval-2016 as a semester project, aiming to build a model for analyzing and classify- ing “tweets” into positive, neutral and nega- tive, according to the evoked sentiment, while getting familiar with Natural Lan- guage Processing tools and methods. There- fore, the backbone of our sentiment analyz- er consists in several off-the-shelf, freely available resources, enhanced with a classi- fier trained on the SemEval-2016 data. 1 Introduction Texts live around us just as we live around them. At any instant, there are texts that people write, share, use to get informed, etc. (starting with an advertisement heard on the radio every morning and finishing with the contract of sale signed be- fore a notary). Combining this with the concept of language economy (or the principle of least ef- fort) – a tendency shared by all humans – con- sisting in minimizing the needed amount of effort to achieve the maximum result, it is no wonder why the social media, with its short, informal and context-dependent texts, achieved such a high popularity. SemEval-2016 task 4 had several subtasks, but since our team consists mainly of members be- ginning to learn about Natural Language Pro- cessing, we only felt comfortable in participating in Subtask A. This subtask involved the classifi- cation of a message polarity, i.e. classify a given tweet in three categories: positive, negative or neutral, according to the identified sentiment (Nakov et al., 2016). The remaining of this paper is structured as follows: Section 2 provides an overview of avail- able online applications for analyzing social me- dia sentiments; Section 3 presents our own sen- timent analyzer, before the final section present- ing the evaluation of the system and drawing some conclusions. 2 State of the art Specific processing tools (such as POS taggers or anaphora resolution systems), score a higher per- formance if used on the same text type as the ones they were trained on. In other words, we will have better results if using a POS tagger trained on news corpora to analyze news texts, rather than speech transcripts. Thus, the short dimension of tweets and their creative informal spelling have raised a new set of challenges to the natural language processing field. How to handle such challenges while au- tomatically mining and understanding the opin- ions and sentiments that people are communi- cating has been the subject of several research (Jansen et al., 2009; Barbosa and Feng, 2010; Bifet and Frank, 2010; Davidov et al., 2010; O’Connor et al., 2010; Pak and Paroubek, 2010; Tumasjan et al., 2010; Kouloumpis …

Content truncated to 500 words