Soft Computing Techniques for Sentiment Analysis and Feature Selection

Understanding the significance of social media has attracted academic attention in recent times. As a prominent scholar once put it, social media is no longer a passing sensation or fad. Customer opinions expressed on social media can convey important messages that businesses can use to build strong relationships with customers. As social media usage among the general population grows, so are its uses in the business world as more businesses turn to social media as a costeffective and efficient way to connect with many clients. Despite the quick transition from traditional to social media, firms still struggle to fully comprehend the needs and concerns of their customers in this era of the so-called big data. Moreso, the ability to quickly comprehend consumer communications so that management can respond in a timely and effective manner remains a key challenge. Further, the huge amount of unstructured data and a scarcity of practical tools for analysing this unstructured data makes such analysis more complicated. This dissertation presents a brief overview of the application of soft computing techniques for sentiment analysis and feature selection. Initially, the author of the dissertation utilizes the abundance of social media data available online as leverage by employing text mining techniques to analyze user-generated content from social media posts (tweets) to support consumer decision-making and marketing communications. This unstructured user-generated content heavily includes slang, misspelt words, etc... thereby presenting a challenge to feature selection due to the vagueness, imprecision, and ambiguity contained therein. Consequently, a metaheuristic-based solution using the Particle Swarm Optimization (PSO) algorithm for optimal text feature selection during sentiment analysis is implemented to enhance sentiment prediction accuracy. The second segment of the dissertation combines evolutionary computation techniques with angle modulation to solve feature selection problems. Eighteen classical UCI machine learning datasets are employed in evaluating the performance of the proposed technique. The authors’ findings confirm the competitive and superior performance of the proposed approach when juxtaposed with other work-related metaheuristics methods available in feature selection literature. Further statistical tests also confirm the proposed method as a potent tool for resolving binary optimization problems across different domains.

ISBN:	978-80-7678-193-1
EAN:	9788076781931
Počet stran	34 stran
Datum vydání	11. 10. 2023
Pořadí vydání	První
Jazyk	anglický
Vazba	E-kniha
Autor:	Raphael Kwaku Botchway
Nakladatelství	Univerzita Tomáše Bati ve Zlíně
Tématická skupina	5 - Technické vědy
	Neprodejná publikace. Publikaci je možné poptávat zde: Volně dostupné na http://hdl.handle.net/10563/52452