Enhanced Medical Tweet Opinion Mining using Improved Dolphin Echolocation Algorithm Based Feature Selection

Abstract

Extraction and analysis of public opinions from social network data can provide interesting outcomes and inferences about product, service, event or personality. Twitter is one of the most popular medium for analyzing the public sentiment through user tweets. Feature specific opinion analysis provides highly accurate and effective classification and categorization of public opinions. This paper focuses on developing an opinion mining framework for automated analysis of tweet opinions using efficient feature selection and classification algorithms. For this purpose, an Improved Dolphin Echolocation Algorithm (IDEA) is developed by enhancing the optimization performance of the Dolphin Echolocation Algorithm (DEA). The limitations of DEA are the insufficient exploration and exploitation properties in local optimum solutions and also impact the convergence rate. These shortcomings are overcome by the proposed IDEA algorithm. In this work, first the tweets are collected and pre-processed to extract the features using Part-of-Speech (POS) tagging and n-grams aided by a dictionary. Using IDEA, the feature subset candidates are selected and the outcomes are fed as input to the baseline classifiers to obtain highly accurate opinion classification. The evaluation of the k-Nearest Neighbor (KNN), Naïve Bayes (NB) and Support Vector Machine (SVM) classifiers using the two feature selection approaches of DEA and IDEA are performed over cancer and drug tweets datasets and the results illustrate that the classification accuracy of opinions is enhanced significantly through the IDEA based feature selection than the traditional DEA algorithm. These results justify the utilization of the proposed IDEA algorithm for improving the opinion mining applications in different fields.