This article studies the task of sentiment detection toward aspects of economic and social development in Russian sentences. The aspect toward which the sentiment is determined can be either explicitly mentioned or implied. The authors study the possibilities of using neural network classifiers and propose an algorithm for determining the sentiment toward an aspect based on semantic rules implemented with the use of constituency trees. The sentiment toward an aspect is determined in two stages. At the first stage, aspect terms (explicitly mentioned events or phenomena associated with the aspect) are found in the sentence. At the second stage, the sentiment toward an aspect is calculated as the sentiment toward the aspect term that is most closely associated with the aspect. This paper proposes several methods for searching the aspect terms. The model’s performance is assessed on a corpus of 468 sentences extracted from election campaign materials. The best result for neural network classifiers is obtained using the BERT-SPC neural network pretrained on the task of identifying the sentiment toward an explicitly mentioned aspect; the macro F-measure is 0.74. The best result for the semantic rule-based algorithm is obtained using the method of aspect term searching based on semantic similarity; the macro-F-measure is 0.63. When combining BERT-SPC and the rule-based algorithm into an ensemble, the macro-F-measure is 0.79, which is the best result obtained in this study.
Poletaev et al. (Mon,) studied this question.