Sentiment Analysis of Polish Texts

A new language resource for sentiment analysis(SA) and an application of SA to a new domain—discussions on an online Polish news forum—are developed. A scheme for human annotation of textual samples is proposed using online questionnaires. A method for classifying the samples based on the annotations is introduced and put into practice. A method, applying existing advanced Information Retrieval (IR)techniques, for SA within a Bayesian learning framework is explored. Preliminary experimental results show the IR techniques, in conjunction with the Naïve Bayes classifier, can be expected to produce good sentiment classification performance both for Polish texts and for the news discussion domain.