Point-wise Mutual Information (PMI) based Weighting Scheme for Sentiment Analysis

Point-wise Mutual Information (PMI) is proved to be very important for Sentiment Analysis. Turney (2002) defined point-wise mutual information as it is computed by the difference of the mutual information between the feature and the word “excellent” and mutual information between the feature and the word “poor”.
We can compute the PMI as the difference of association each word with positive class and negative class. The computed PMI value can be used as the weights for the words in the vector space model instead of the weights given by TF-IDF, TF and binary weighting methods. Further, vector space model can be built with the given PMI values. Finally, machine learning models van be built with the vector space model built and performance can be compared within TF-IDF, TF, Binary and PMI based weighting scheme for sentiment analysis.

No Comments

This post is password protected. Enter the password to view any comments.

Post a Comment