Learning Based Methods for Addressing Exceptions in Sentiment Analysis

Abstract

Sentiment analysis detects how users feel about speciand#64257;c emotional issues fre- newlinequently discussed on social media. Machine learning approaches are often newlinefavored for the same reason. On the other hand, supervised approaches en- newlinetail labeled data, which isn t always available. On the other hand, cluster- newlineing approaches are quite eand#64256;ective for sentiment analysis. As a result, this newlineresearch focuses on developing eand#64259;cient clustering algorithms for sentiment newlineanalysis. For sentiment analysis tasks, partitional clustering algorithms out- newlineperform other clustering methods in the literature. They do, however, have newlinerestrictions, such as prior knowledge of the number of clusters, the production newlineof local solutions, and so on. As a result, metaheuristic approaches are used newlinein partitional clustering; however, their convergence precision is weak. newlineIn this report, two important contributions try to build eand#64256;ective metaheuristic newlineclustering approaches to address such challenges. To begin, two metaheuristic newlinestrategies have been proposed to improve convergence precision: grey wolf op- newlinetimization with k-Means (GWOK) clustering method and and#64257;tness-based grey newlinewolf optimization with k-Means (FGWOK) clustering method solve the prob- newlinelem of spam detection. To eliminate the irrelevant and redundant features, the newlineChi-square feature selection method has been used. A comparative analysis newlineof other feature selection methods such as Forward Selection, Decision-Tree newlinebased selection, and ANOVA are presented in this report. The experimen- newlinetal analysis shows that the proposed methods outperform the other existing newlinemethods. newline