SentiTFIDF – Sentiment Classification using Relative Term Frequency Inverse Document Frequency

Sentiment Classification refers to the computational techniques for classifying whether the sentiments of text are positive or negative. Statistical Techniques based on Term Presence and Term Frequency, using Support Vector Machine are popularly used for Sentiment Classification. This paper presents an approach for classifying a term as positive or negative based on its proportional frequency count distribution and proportional presence count distribution across positively tagged documents in comparison with negatively tagged documents. Our approach is based on term weighting techniques that are used for information retrieval and sentiment classification. It differs significantly from these traditional methods due to our model of logarithmic differential term frequency and term presence distribution for sentiment classification. Terms with nearly equal distribution in positively tagged documents and negatively tagged documents were classified as a Senti-stop-word and discarded. The proportional distribution of a term to be classified as Senti-stop-word was determined experimentally. We evaluated the SentiTFIDF model by comparing it with state of art techniques for sentiment classification using the movie dataset.