SNS 해시태그를 이용한 사용자 감정 분류 방법에 관한 연구

Abstract

Recently, studies are being actively carried out for sentiment analysis, which is a type of natural language processing technologies for analyzing subjective data such as opinions, attitudes, and propensities of users expressed on Web, blogs, and social network services (SNSs). Conventionally, to classify the sentiments of texts, most studies were carried out in a form of determining positive/negative/neutral sentiment by assigning polarity values for sentiment vocabulary by using a sentiment Lexicon. However, in this study, the sentiments were classified on the basis of Thayer's model defined psychologically, unlike the polarity classification used in the opinion mining.

In this paper, as a method of classifying the sentiments, sentiment categories were proposed by extracting sentiment keywords for major sentiments by using hashtags, which are essential elements of Instagram. By applying this to user posts, sentiments can be determined through similarity measurement between the sentiment adjective candidates and sentiment category’s sentiment keywords. Also, since sentiment distribution and major sentiments of users can be determined, the ambiguity of subject sentiments can be solved objectively. As a test result for the proposed method, the average accuracy rate for the whole sentiment categories was 90.7%, showing a good performance.

In this study, the sentiment categories were proposed by selecting happy, angry, peaceful, and sad as categories as typical sentiments to improve the accuracy of sentiment classification and minimize the misclassification rate; but in the future, if classifiable sentiments are additionally selected and expended on top of the four sentiments, it is thought that user sentiments can be analyzed more specifically. Furthermore, this study was carried out only with sentiment adjectives extracted by using texts, but it should be extended to a study including a method of using emoticons or other parts of speech containing sentiments.

Through the proposed method, if a sentiment classification system of large capacity is prepared, it is expected that sentiment analysis will be possible for various fields such as major issues and social phenomena through SNS, and furthermore, it is expected to be used in user tailored services, recommendation services, or sentiment marketing on SNS.