Semantic rule-based sentiment detection algorithm for Russian publicism sentences

Abstract

The article is devoted to the task of sentiment detecton of Russian sentences, which is understood as the author’s attitude on the sentence topic expressed through linguistic expression features. Today most studies on this subject utilize texts of colloquial style, limiting the applicability of their results to other styles of speech, particularly to the publicism. To fill the gap, the authors developed a novel publisism sentences oriented sentiment detection algorithm. The algorithm recursively applies appropriate rules to sentence parts represented as constituency trees. Most of the rules were proposed by a philology expert, based on knowledge on the expression features from Russian philology, and algorithmized using constituency trees generated by the algorithm. A decision tree and a sentiment vocabulary are also used in the work. The article contains the results of evaluation of the algorithm on the publicism sentences corpus OpenSentimentCorpus, F-measure is 0.80. The results of errors analysis are also presented.