Polarity Detection in Blog Comments from Blog Rss Feed by Modified TF - IDF Algorithm

412 | P a g e ABSTRACT Blogs are most common medium over web where user posts their opinion. It is considered to be a web space of the users where they share their views, beliefs and other philosophy. Blogs posted across the web can be extracted from their rss feed. Once a blog is posted, several readers leaves their comment on the blogs. Analyzing these comments can help in finding the opinion of people for the blog or about the topic in general. A general pattern for these comments are, they are short and sometimes not very grammatically accurate. Many a times the comments are generalized like an appreciative statement for the author or about the post. There are several opinion polarity mining techniques which are mainly eccentric around the theory of training a natural language processing machine with known opinionistic blogs and train a classifier based on this. The classifier further classifies the blogs based on their closeness with the trained datasets. A machine learning in natural language processing requires huge training data to build the decision rule and therefore classification time also increases naturally. Therefore this work proposes a unique technique of opinion polarity mining from comments of the blogs dynamically, through the RSS feed with a unsupervised classifier. The proposed technique is based on modified TF-IDF algorithm for first extracting the relevance of a comment with the topic and thereafter uses a scoring mechanism to identify the opinion based on occurance of opinionisitic terms and their order in comments. The algorithm is tested with various real time blog site like digitalinspiration.com, techmafia.org, integratedideas.co.in, kerryseo.co.in and so on. RSS of blogger,wordpress and blogspot powered blogs are tested for testing the efficiency of the detection. 100 posts in total are analyzed and are verified by the author of the posts about effectiveness of the opinion being detected. Result shows an overall accuracy of 81% in classifying the opinion.