TEXT SENTIMENT ANALYSIS BASED ON FUZZY RULES AND INTENSITY MODIFIERS
Abstract
Expressing feelings is a hidden part of hard life and communication. To create computers that can better serve humanity, computer science continues to research into developing machine learning algorithms that can process text data and perform sentiment analysis tasks on natural language texts. Additionally, the availability of online reviews and increased end-user expectations are driving the development of system intelligence that can automatically categorize and share user reviews. Every year, research in this area has discovered more and more emotions in text, but only a small part of it has been devoted to the use of fuzzy logic. This mainly happens because the researchers often use binary classification – «positive » and «negative», less often adding a third class – «neutral». The use of fuzzy logic helps to determine emotions, and not just «good» and «bad», but the degree of these emotions. The number of classes is defined by determines of the level of detail. Previously, we proposed a fuzzy dictionary-based sentiment model, in this paper we propose an improved text sentiment determination model based on a sentiment dictionary (SentiWordNet) and fuzzy rules. To determine the accuracy and precision of sentiment analysis, coefficients were applied to observe the emotional load of words of different parts of speech and action modifiers that contribute to the strengthening or weakening of emotional tones. The quantitative value of the sentiment of the text is obtained by aggregating normalized data by emotional classes using fuzzy result methods. As a result of the study, it was found that taking into account all modifiers can significantly increase the accuracy of the method previously proposed by the authors, and also ensures the determination of boundaries when determining a detailed assessment of relationships in 7 classes (“very positive”, “positive”, “somewhat positive”, “neutral” , “somewhat negative”, “negative”, “very negative”).
