Abstract: The social media platform is important as a means to spread information, opinions or awareness about real events. Twitter is prominent because of the huge amount of messages on all kinds of topics posted every day. Text classification of Twitter is never a trivial task that can be handled by common natural language processing techniques. There is also no agreement as to the definition of what type of operation will be performed in the event identification and classification of Tweets. Existing methods are difficult to reproduce and compare because they are often focused on certain types of events according to certain assumptions. This approach is in case of distinct nature. Previous work neglects impactful words due to the use of Bag-of-Words model. The contribution is for feature selection, Information Gain (IG) with consideration of adjectives as most impactful feature in POS tagger is used. Similarity Model, Relevance, Specificity & Span of relationship computation is also carried out later. SVM classification approach is implemented to construct positive and negative sentiments terms. Further this trained dataset is used for SVM classification of test dataset. Results show promising improvements.
Keywords: Feature Selection, Information Gain, Support Vector Machine, Sentiment Classification, Sentiment Analysis
| DOI: 10.17148/IJARCCE.2018.7815