Feature Engineering for Chinese Microblog Sentiment Classification
LI Zeku · Journal of Shanxi University · 2014
Sentiment classification,a basic sentiment analysis task,aims to classify a sentiment sentence into positive,negative and neutral.Sentiment analysis on microblog is challenging,which is different from it on common product reviews,due to the characteristics of microblog.Many previous works used machine learning based approaches to solve this task,the core of which is to try and select useful features,for instance,bag of words.However,these proposed features may not be suitable for Chinese due to linguistic differences.What is more,there is no feature engineering for Chinese microblog in details.In this paper,we do some feature engineering for Chinese microblog sentiment classification,from words,phrases,numbers,syntactic features,and new feature named dictionary-rule based sentiment score,in order to make a better performance beyond the baseline.At last,we obtain reliable feature set through a large number of experiments and analysis.Our approach significantly improves the results of sentiment classification.