- Model introduction
Based on the theory of sentiment classification, the sentiment analysis model analyzes the basic sentiment vocabulary of social media (include not only standard sentiment vocabulary, but also emotional expression terms commonly used on social media), and combines online text vocabulary matching technology to analyze and measure the intensity distribution of five basic social emotions: happiness, sadness, anger, fear and disgust.
2. R&D Bases
Dong Yinghong, Chen Hao, Lai Kaisheng, Yue Guoan. Weibo Social Moods Measurement and Validation. Psychological Science.2015,38(5):1141-1146.
3. Algorithm Description
(1) Construction of the basic sentiment corpus of social media
With reference to the documentation method of the mood scale POMS, a POMS-1 suitable for Twitter research was constructed (Pepe & Bollen, 2008). First 65 words of the POMS scale were used as seed words, and their synonyms from Wordnet (3rd edition) and “Roger New Millennium Synonyms Dictionary (1st edition)”, plus the 1500 nouns used by Wang, Zhou and Luo(2008) in their study of the sentiment dimension, and finally expands into the sentiment corpus with some additions, deletions, and classifications by professional research scholars.
(2) The calculation process of the five basic social emotions is: identify all the emotional words in the post through the sentiment corpus, and calculate the score of the post on various emotional types based on the weight of the post, the higher the score, the stronger the mood, and vice versa. The model identifies the sentiment category with the highest score as the emotion of the post by default.
4. Applicable scenarios
(1) For natural language texts with obvious subjective emotions, the reference value of automatic analysis is more effective.
(2) For natural language texts that are not particularly complex in sentence structure, the reference value of automatic analysis is more effective.