Identification of emotions in text articles through data pre-processing and data mining techniques

Automatic identification of emotions from text has been an interesting area of research in the recent times. With the exponentially growing number of online users, emotion identification of the text being read by the users can help in providing a comforting environment to the readers. In this paper, emotion classification is attempted on the news articles from the society channel of Sina. The proposed methodology comprises of data pre-processing followed by application of data mining on the pre-processed data. The data pre-processing step involves conversion of Chinese to English, removal of html tags, removal of stop words and stemming. The feature vector to be provided for data mining is then formulated through the identified root words and the dictionary of terms from the WordNet-Affect. Since the dimensionality of the feature vector thus formed is very high, feature selection procedure is adopted to attain the most significant features. Various classification algorithms were attempted on the reduced feature set to classify the emotions as positive, negative, mixed or neutral. Functional Trees with Bagging provided the best possible results achieving an accuracy of 88.41%.