A hybrid approach for building a Vietnamese sentiment dictionary

Sentiment (or emotional) dictionaries play a key role and are “the materials” for sentiment analysis applications, especially in Vietnamese where the corpus is not yet available. In this research, we propose a sentiment dictionary for Vietnamese, known as VNSD, which adopts machine translation, Logistic Regression, “double propagation,” and fuzzy rules. The sentiment dictionary counts approximately 5,000 adjectives, 2,000 verbs, 300 nouns, and more than 200 adverbs, which lead to a combination of more than 100,000 sentiment phrases covering the Vietnamese emotional lexicons. We found that a hybrid approach offers better performance compared to stand-alone data-mining or natural language processing techniques. The key contribution of VNSD that distinguishes it from related work is a deep investment in exploiting Vietnamese linguistic characteristics to propose suitable rules for computing the sentiment scores of Vietnamese text phrases.