Can Korean Functional Suffixes Trigger Sentiment in Short Reviews?

다국어 초록 (Multilingual Abstract)

Grammatical categories contributed to sentiment analysis have been mostly a word or a bag of words which express semantic orientation. Korean, however, presents some of the sentence-final suffixes to be exploited for identifying sentiment. Those suffixes mark honorification which are interpreted as lowered or raised speech depending on their type. This research has referred them as Sentiment-Triggering Suffixes (STSs) and extracted them from a large sentiment corpus, utilizing Chi-square test and Association rules. A series of experiments has proved that using STSs as a feature is effective in detecting sentiment particularly when a sentence is short in length and does not contain a sentiment-bearing word. In addition, better results have been derived when STSs imposing negative sentiment were implemented. Utilizing STSs, in particular, helps identify rhetorical question, which is considered as a challenging task. Since cutting-edge machine learning techniques can visualize the most important features of a classifier, we have illustrated the characteristics of STSs using two of the techniques, t-SNE and heatmap.

목차 (Table of Contents)

• 1. Introduction
• 2. Related Studies
• 3. Extracting Sentiment-Triggering Suffixes
• 4. Defining Characteristics of Sentiment-Triggering Suffixes
• 5. Conclusion
• References

참고문헌 (Reference)

1 김지은, "설명자의 영화평 감성 분석 결과 해석에 대한 고찰" 한국언어학회 45 (45): 797-820, 2020

2 Ribeiro, Marco Tulio, "Why should I trust you?: Explaining the predictions of any classifier" 1135-1144, 2016

3 Van der Maaten, Laurens, "Visualizing high dimensional data using t-SNE" 9 : 2579-2605, 2008

4 Sadock, Jerrold M, "Toward a Linguistic Theory of Speech Acts" Academic Press 1974

5 Cho, Danbi, "Subword-based sentence representation model for sentiment classification" 2020

6 Liu, Bing, "Sentiment Analysis: Mining Opinions, Sentiments, and Emotions" Cambridge University Press 2015

7 Pedregosa, Fabian, "Scikit-learn: 16 Machine learning in python" 12 : 2825-2830, 2011

8 Sadock, Jerrold M, "Queclaratives" 223-232, 1971

9 Lee, Sangah, "KR-BERT: A small-scale korean-specific language model" abs/2008.03979 : 2020

10 Han, Chung Hye, "Interpreting interrogatives as rhetorical questions" 112 : 201-229, 2002

11 Friedman, Jerome H, "Greedy function approximation: A gradient boosting machine" 29 : 1189-1232, 2000

12 Piatetsky-Shapiro, Gregory, "Discovery, analysis, and presentation of strong rules" 1991

13 Bagui, Sikha, "Deriving strong association mining rules using a dependency criterion, the lift measure" 1 (1): 297-312, 2009

14 Park, Kwang-Hyeon, "Bert for Korean natural language processing: Named entity tagging, sentiment analysis, dependency parsing and semantic role labeling" 584-586, 2019

15 Cho, Danbi, "An empirical study of korean sentence representation with various tokenizations" 10 (10): 2021

16 Meesad, Phayung, "A chisquare-test for word importance differentiation in text classification" 110-114, 2011