Effective Emoticon Based Framework for Sentimental Analysis of Web Data

Abstract

The Explosive development in the social media domain has created a platform for mass generation of textual and emoticon based web data from micro blogging sites. Sentimental Analysis refers to analysis of sentiments or emotions from such heterogeneous reviews are the present urge of the market. Thus, an effective emoticon based framework is proposed which generates scores of both textual and emoticons into seven layered categories using SentiWordNet and weighs performance of various machine learning techniques like SVM/SMO, K-Nearest Neighbor (IBK), Multilayer Perception (MLP) and Naive Bayes (NB). Using Jsoup crawler input reviews are obtained and processed with initial pre-processing model for emoticons and text data followed by stemming and POS tagger. Projected framework is investigated on college and hospital dataset obtaining upper attainment level by Kappa statistic metrics having 98.4% correctness and lesses bug value. Proposed Framework showcases greater competence score with lesser FP Rate based on weighted average of correctness measures. The investigational outcomes are tested on training data with Ten-Fold cross validation. The outcome reveals that suggested emoticon based framework for the task of Sentimental analysis can be efficaciously applied in online decision job.