UMCC_DLSI: A Probabilistic Automata for Aspect Based Sentiment Analysis

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 722–726, Dublin, Ireland, August 23-24, 2014. UMCC_DLSI_Prob: A Probabilistic Automata for Aspect Based Sentiment Analysis Yenier Castañeda Armando Collazo Elvis Crego Jorge L. Garcia Yoan Gutiérrez David Tomás Andrés Montoyo Rafael Muñoz DI, University of Matanzas Matanzas, Cuba DLSI, University of Alicante Alicante, Spain {yenier.castaneda, armando.collazo}@umcc.cu, elvis.crego@mtz.cu, jorge.garcia@infonet.umcc.cu {ygutierrez,dtomas,montoyo, rafael}@dlsi.ua.es Abstract This work introduces a new approach for aspect based sentiment analysis task. Its main purpose is to automatically assign the correct polarity for the aspect term in a phrase. It is a probabilistic automata where each state consists of all the nouns, adjectives, verbs and adverbs found in an annotated corpora. Each one of them contains the number of occurrences in the annotated corpora for the four required polarities (i.e. positive, negative, neutral and conflict). Also, the transitions between states have been taken into account. These values were used to assign the predicted polarity when a pattern was found in a sentence; if a pattern cannot be applied, the probabilities of the polarities between states were computed in order to predict the right polarity. The system achieved results around 66% and 57% of recall for the restaurant and laptop domain respectively. 1 Introduction Sentiment analysis is increasingly viewed as a vital task from both an academic and a commercial standpoint. Textual information has become one of the most important sources of data to extract useful and heterogeneous knowledge. “Texts can provide factual information, such as: descriptions, lists of characteristics, or even instructions to opinion-based information, which would include reviews, emotions, or feelings. These facts have motivated dealing with the identification and extraction of opinions and sentiments in texts that require special attention.” (Gutiérrez, et al., 2014). Sentiment Analysis or “Subjectivity Analysis” in (Liu, 2010) is defined as the computational treatment of opinions, sentiments and emotions expressed in a text. In order to automatically treat the subjectivity, we need lexical resources that allow the detection and evaluation of the affective/ subjective charges in texts, its polarity and intensity. Regarding research carried out for linguistic patterns identification and its polarity in texts, it is worth mentioning works on: adjectives (Hatzivassiloglou and McKeown, 1997) (Wiebe, 2000); adjectives and verbs (Turney, 2002) (Wilson, et al., 2005) (Takamura, et al., 2007); and also verbs and names (Esuli and Sebastini, 2006). WordNet (Fellbaum, 1998) has also been used for the collection of opinion adjectives and verbs (Kim and Hovy, 2005) to determine the semantic orientation of the terms depending on their notes (Esuli and Sebastiani, 2005), for the adjective extraction (Andreevskaia and Bergler, 2006) or opinion mining (Esuli and Sebastiani, 2007). Inspired on Hidden Markov models (Baum and Petrie, 1966) and following the idea that words combinations are finite in an evaluation text, we decided to create a finite automata in graph form to represent all these relations extracted from a training corpus. For the creation of this automata we utilised different resources, such as WordNet and OpinionFinder Subjectivity Lexicon. Also, different extracted …

Content truncated to 500 words