Classifier-based Polarity Propagation in a WordNet

Classifier-based Polarity Propagation in a Wordnet Jan Koco´n, Arkadiusz Janz, Maciej Piasecki Faculty of Computer Science and Management Wrocław University of Science and Technology, Wrocław, Poland {jan.kocon, arkadiusz.janz, maciej.piasecki}@pwr.edu.pl Abstract In this paper we present a novel approach to the construction of an extensive, sense-level sentiment lexicon built on the basis of a wordnet. The main aim of this work is to create a high-quality sentiment lexicon in a partially automated way. We propose a method called Classifier-based Polarity Propagation, which utilises a very rich set of wordnet-based features, to recognize and assign specific sentiment polarity values to wordnet senses. We have demonstrated that in comparison to the existing rule-base solutions using specific, narrow set of semantic relations, our method allows for the construction of a more reliable sentiment lexicon, starting with the same seed of annotated synsets. Keywords: sentiment analysis, polarity propagation, wordnet 1. Introduction Sentiment analysis of natural language utterances is con- tinuously increasing its importance as one of the most ex- pected techniques. The best results have been obtained with supervised approaches trained on the basis of anno- tated texts from a selected domain, e.g. movie or product reviews. However, cross-domain applications show a sig- nificant drop in the performance of classifiers trained on a corpus built from texts from the other domains. This can be attributed to a high correlation of a classifier with words and phrases that are specific for the positive and negative utterances of the given domain. However, language ex- presses some lexical means of conveying sentiment polarity in a way that is shared across different domains. A com- promise between performance and domain adaptability can be achieved using hybrid methods. A lexicon of sentiment polarity could be a very useful basis for constructing such a domain independent, hybrid system, if such a lexicon is large, comprehensive and reliable enough. plWordNet 3.1 emo1 (Maziarz et al., 2016; Za´sko-Zieli´nska and Piasecki, 2018) is a very large lexical semantic net- work for Polish, in which more than 190,000 different lem- mas and 285,000 Polish lexical meanings are described by the lexico-semantic relations. plWordNet has become one of the largest Polish dictionaries ever built, and the largest wordnet in the world. A substantial part of plWord- Net was manually described by emotive annotation (Za´sko- Zieli´nska and Piasecki, 2018). In this work we would like to expand this emotive annotation to a very large scale and make plWordNet a basis for a large hybrid emotive lexicon of Polish, as well as for the development of a hybrid sys- tem for sentiment and emotion analysis in Polish texts. Ex- isting solutions for sentiment propagation over a wordnet are mostly based on a simple label propagation algorithm, starting with a relatively small initial seed. However, gen- erally they do not take into account more complex word- net structures, thus we may lose some information which can be a good indicator of sentiment polarity. Exploiting the full wordnet structure gives us an opportunity to cap- 1 http://plwordnet.pwr.edu.pl ture the polarity …

Content truncated to 500 words