SEMANTIC-STATISTICAL ALGORITHM FOR DETERMINING THE CATEGORIES OF ASPECTS IN THE PROBLEMS OF SENTIMENT ANALYSIS

Abstract

In the modern world, one of the important communication channels is the Internet. Trade, promotion of services is carried out through electronic platforms. Social networks and instant messengers are becoming the most important communication channel and a powerful tool for influencing public opinion. A significant amount in all published content falls on texts written in natural language. Therefore, the problems of natural language processing (NLP) and natural language understanding (NLU) today are one of the key ones. Under the influence of commercial interests, the field of automatic aspect-based sentiment analysis is actively developing. This task significantly depends on specific subject areas, and therefore the issue of quick and effective adaptation of existing models to new domains is very acute. The paper proposes a hybrid method of aspect-oriented analysis, based on data extracted from common dictionaries and domain-oriented texts. The novel method for constructing a condensed semantic graph based on unstructured domain- dependent texts is proposed. Numerical metrics to assess the significance of individual termswithin the entire domain are introduced. An algorithm for the text categorization based on the selection of semantic clusters within a condensed domain-specific graph is proposed. A method for assessing the sentiment of domain-oriented texts based on statistical data, including the joint use of a tone lexicon and a condensed domain-specialized graph, is proposed. The results of experiments are presented, allowing for evaluation of the quality of the algorithms.