The sentiment dictionary plays an important role in analyzing or identifying opinion of users. A Sentiment dictionary is widely applicable to many different domains. Therefore, many researchers are interested in and building sentiment dictionaries. However, most of these dictionaries were built based on Vietnamese lexicon, when applied to social reviews often have low accuracy, because of the way social media is used different from Vietnamese lexicon. In this paper, we present a methodology of constructing Vietnamese sentiment dictionary with scoring for analyzing opinion of social reviews. We used training set with 5,200 labeled sentences that were collected from customer’s reviews about electronic product domain on electronic commerce websites. After that, we extracted nouns, adverbs and adjectives and then applied support measurement to calculate weight of them. The experimental results with 249 sentences have an accuracy of approximately 92% compared to 87% of dictionary that is developed based on Vietnamese lexicon, showing that our dictionary has a higher accuracy when applied to social sentiment analysis.