VietSentiLex: a sentiment dictionary that considers the polarity of ambiguous sentiment words

VietSentiLex: a sentiment dictionary that considers the polarity of ambiguous sentiment words Vo Huynh Quoc Viet Nagaoka University of Technology Natural Language Processing Lab 1603-1 Kamitomioka, Nagaoka, Niigata, Japan viet@jnlp.org Kazuhide Yamamoto Nagaoka University of Technology Natural Language Processing Lab 1603-1 Kamitomioka, Nagaoka, Niigata, Japan yamamoto@jnlp.org Abstract The ability to analyze sentiment is a ma- jor technology to analyze social media pro- cess. The sentiment analysis involves read- ing and understanding what is being said about a brand as well as advertising cam- paigns in online services to determine the nature of a product. Because the Viet- namese language has few resources for ap- plying machine learning tasks, use of sen- timent dictionaries is required. In this study, a sentiment dictionary called ”Viet- SentiLex” is introduced for the aforemen- tioned task in the Vietnamese language. Most notably, , In this study, instead of applying scores for every word, ambiguous words are considered more carefully as it is periodically positive or negative. Related words such as target nouns or verbs are used as contextual information for a senti- ment word. Experiments to compare the performance of our dictionary to others are conducted. We prove that our dictionary has a high potential in predicting the po- larity of reviews as compared to other dic- tionaries. In addition, various challenges and disadvantages of this dictionary are also outlined for future improvement until VietSentiLex can be a commercial prod- uct. Keywords: Sentiment Analysis, Dictio- nary, Ambiguous Words 1 Introduction Sentiment analysis is one of the specified smaller tasks of opinion mining, which is sub- discipline at the crossroads of information re- trieval and computational linguistics. It involves the contextual mining of sentences to identify sentiment information in input material. Fur- thermore, it summarizes the target into speci- fied classes such as positive, negative, neutral, interesting, or bad. Thus, sentimental analysis helps a business to roughly estimate user opin- ions of their brand, products, or services using online reviews and conversations. With the increase in online comments and re- views on social networks every day, sentiment analysis has begun to play an increasingly more important role in automatically evaluating user opinion, which is manually impossible for a busi- ness. Recently, the application of deep learning for this kind of task has become quite common. However, it requires both large and highly ac- curate sets of data for training, which is reason- ably impossible to obtain with low-resource lan- guages like Vietnamese. Thus, in this study, the traditional dictionary-based approach is consid- ered by building a new sentiment dictionary in the Vietnamese language. This study also introduces a new approach to classify sentiment polarity. All previous works used normal sentiment classification terms such as ”positive,” ”negative,” ”neutral,” or more specifically, ”sad,” ”fun,” ”happy.” Some of the most popular dictionaries such as SentiWord- Net of Andrea and Fabrizio (2006) use the terms ”positive,” ”negative,” and ”objective.” Another is EmoLex by Saif and Peter (2014), which includes the terms ”positive,” ”negative,” and PACLIC 32 706 32nd Pacific Asia …

Content truncated to 500 words