UCSC-NLP at SemEval-2017 Task 4: Sense n-grams for Sentiment Analysis in Twitter

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 807–811, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics UCSC-NLP at SemEval-2017 Task 4: Sense n-grams for Sentiment Analysis in Twitter Jos´e Abreu, Iv´an Castro, Claudia Mart´ınez, Sebasti´an Oliva Catholic University of the Most Holy Conception Alonso de Ribera 2850, Concepci´on, Chile {joseabreu,cmartinez}@ucsc.cl {ilcastro,snoliva}@ing.ucsc.cl Yoan Guti´errez University of Alicante Carretera de San Vicente s/n Alicante, Spain ygutierrez@dlsi.ua.es Abstract This paper describes the system submit- ted to SemEval-2017 Task 4-A Senti- ment Analysis in Twitter developed by the UCSC-NLP team. We studied how relationships between sense n-grams and sentiment polarities can contribute to this task, i.e. co-occurrences of WordNet senses in the tweet, and the polarity. Fur- thermore, we evaluated the effect of dis- carding a large set of features based on char-grams reported in preceding works. Based on these elements, we developed a SVM system, which exploring SentiWord- Net as a polarity lexicon. It achieves an F1 = 0.624 of average. Among 39 sub- missions to this task, we ranked 10th. 1 Introduction To determine whether a text expresses a POSI- TIVE, NEGATIVE or NEUTRAL opinion has at- tracted an increasingly attention. In particular, sentiment classification of tweets has immediate applications in areas such as marketing, political, and social analysis (Nakov et al., 2016) Different approaches have shown to be very promising for polarity classification of tweets such as Convolutional Neural Networks trained with large amounts of data (Deriu et al., 2016). Several authors have studied Machine Learning approaches based on lexicon, surface and semantic features. The proposal of Mohammad et al. (2013) as well as an improved version of Zhu et al. (2014) show very competitive scores. The latter approach was re-implemented by Ha- gen et al. (2015) as a part of an ensemble of twitter polarity classifier which is top-ranked in the Se- mEval 2015 Task 9: Sentiment Analysis in Twit- ter. Our system proposes to enrich the set of fea- tures used by Mohammad et al. (2013). We de- scribe here only the features more relevant for our experiments, further details in all features could be found in Mohammad et al. (2013); Hagen et al. (2015). • Lexicon Based Features (LB) NRC-Emotion, NRC-Sentiment140, NRC- Hashtag (Mohammad et al., 2013), BingLiu (Hu and Liu, 2004) and MPQA (Wilson et al., 2005) lexicons have been used to generate features. Given a tweet, the following features were computed: - Number of words with positive score - Number of words with negative score - Sum of the positive scores - Sum of the negative scores - Maximum positive score - Minimum negative score - Score of the last positive word - Score of the last negative word For unigrams, bigrams and non-contiguous pairs were computed separated feature sets. • N-gram Based Features (WG and CG) Each 1 to 4-word n-gram present in the training corpus is associated with a feature which indicates if the tweet includes or not the n-gram. For char- acters, all different occurrences …

Content truncated to 500 words