A Comparative Study on Key Technologies of the Chinese Sentiment Classification Preprocessing
Zhu Huiyi · Journal of Intelligence · 2011
This paper is a comparative study of the different combination among the five classical algorithms(DF-Document Frequency,IG-Information Gain,MI-Mutual Information,CHI-χ2 Distributor,WET-Weight of Evidence for Text) and the three common weight computing algorithms(Boolean Weight BW,Term Frequency TF,TF-IDF) in text classification preprocessing.A Support Vector Machine(SVM) was selected as the evaluating classifier.It was found that the combination of IG and TF-IDF had a good performance in the test while the combination of WET and TF had a poor one,and the reasons were analyzed theoretically