Enhancing the Performance of Aspect based Sentiment Analysis and Opinion Summarization using Ensemble and Hybrid Models

Abstract

Significant advancements in e-commerce and Web 2.0 have resulted in the creation newlineof various websites, as well as providing the opportunity for customers to express their newlinethoughts about services and features in the form of reviews. Knowing these opinions and newlinetheir associated feelings are critical in decision-making processes. Unfortunately, newlinereferencing these reviews becomes an almost difficult task due to their abundance. So the newlinecustomer s need to summarize the reviews based on their opinion. Individuals, newlinemanufacturers and organizations will be able to make better decisions with the help of newlineopinion summary. The techniques such as aspect extraction, sentiment analysis and newlineopinion summarization are useful to solve this problem. The aim of this research is to newlinecreate an Aspect-based Sentiment Analysis and Opinion Summarization (ABSAOS) newlinemodel that creates a short, accurate, and concise aspect based opinion summary from a newlinelarge corpus of online reviews. newlineThis research proposes ensemble and hybrid summarization models to enhance the newlineperformance of ABSAOS. The restaurant reviews from the Tripadvisor website is used as newlinethe dataset for this research. This research work consists of three phases: Pre-processing newlineand Aspect Extraction, Sentiment Analysis and Opinion Summarization. The preprocessing newlinestage cleans and makes the dataset ready for processing. The major purpose of newlineaspect extraction is to identify the most relevant aspects within a corpus of text using topic modelling techniques. This phase also extracts aspect-specific reviews and stores them in newlineseparate files. newlineThe sentiment analysis phase extracts and analyzes the opinions expressed in the newlinereview dataset. To enhance the performance of the proposed model, this phase introduces newlinetwo ensemble machine learning techniques. This phase also addresses the problem of class newlineimbalance. The performance of the proposed ensemble models is also compared with the newlineother conventional machine learning classifiers. The opinion summarization phase newlinepresents two hybrid summarization models