A Comparison of Approaches for Sentiment Classification on Lithuanian Internet Comments

摘要

Despite many methods that effectively solve sentiment classification task for such widely used languages as English, there is no clear answer which methods are the most suitable for the languages that are substantially different. In this paper we attempt to solve Internet comments sentiment classification task for Lithuanian, using two classification approaches -knowledge-based and supervised machine learning. We explore an influence of sentiment word dictionaries based on the different parts-of-speech (adjectives, adverbs, nouns, and verbs) for knowledge-based method; different feature types (bag-of-words, lemmas, word n-grams, character n-grams) for machine learning methods; and pre-processing techniques (emoticons replacement with sentiment words, diacritics replacement, etc.) for both approaches. Despite that supervised machine learning methods (Support Vector Machine and Naieve Bayes Multinomial) significantly outperform proposed knowledge-based method all obtained results are above baseline. The best accuracy 0.679 was achieved with Naieve Bayes Multinomial and token unigrams plus bi-grams, when pre-processing involved diacritics replacement.