속성 기반 감성 표현의 구(Phrase) 단위 실현 양상 비교 연구
Phrase-level Realization Patterns of Attribute-based Sentiment Expressions
2
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study examines phrase-level realization patterns of sentiment expressions in attribute-based sentiment analysis by applying statistical measures such as absolute frequency, relative frequency, TF-IDF, and Z-score to IT device review data. The analysis revealed high-frequency sentiment expressions shared across attributes as well as attribute-specific POS n-grams that highlight localized sentiment tendencies. Additionally, sentiment polarity distributions associated with these n-grams demonstrated clear patterns of positive, neutral, and negative expressions, providing empirical evidence of how sentiment is realized structurally. These findings underscore the value of combining frequency- and score-based approaches to systematically identify attribute-specific sentiment phrases and refine phrase-based sentiment analysis in Korean. These findings provide a structured linguistic foundation that can be leveraged in future research to enhance the reasoning and prompt design of large language models (LLMs) for more accurate aspect-based sentiment analysis in Korean.
Abstract
This study examines phrase-level realization patterns of sentiment expressions in attribute-based sentiment analysis by applying statistical measures such as absolute frequency, relative frequency, TF-IDF, and Z-score to IT device review data. The analysis revealed high-frequency sentiment expressions shared across attributes as well as attribute-specific POS n-grams that highlight localized sentiment tendencies. Additionally, sentiment polarity distributions associated with these n-grams demonstrated clear patterns of positive, neutral, and negative expressions, providing empirical evidence of how sentiment is realized structurally. These findings underscore the value of combining frequency- and score-based approaches to systematically identify attribute-specific sentiment phrases and refine phrase-based sentiment analysis in Korean. These findings provide a structured linguistic foundation that can be leveraged in future research to enhance the reasoning and prompt design of large language models (LLMs) for more accurate aspect-based sentiment analysis in Korean.