Aspect-Based Sentiment Analysis for Government Smart Applications Customers’ Reviews
The British University in Dubai (BUiD)
Nowadays, sharing opinions has been made easier with the evolvement of Web 2.0. People can share their opinions on their daily activities and consider others’ opinions to decide whether to buy a product or install an app or use a service. Therefore, the public opinion on the web has become a norm in the modern world. Government agencies and business owners are keen to understand the publics’ opinions towards their services and products. This is a key input for these organizations decision making process in terms of understanding the customers’ needs in order to enhance the product or improve the service or introduce new features. This dissertation presents a holistic review on a variety of recent articles that commences with a background on Sentiment Analysis (SA) as well as it touches on numerous SA techniques, issues, challenges and real-life applications with focus on governmental services and smart apps. In this study, the government smart applications aspects that can be used in aspect-based SA were defined based on written standards with emphasis on customer experience as an important aspect. The proposed aspects include User Interface, User Experience, Functionality and Performance, Security, as well as Support and Updates. For studying SA of government smart applications customers’ reviews, a novel domain-specific annotated dataset has been constructed. It involves government apps in the United Arab Emirates (UAE) as well as its corresponding aspects terms and opinion lexicons. This was done with the help of a proposed Government Apps Reviews Sentiment Analyser (GARSA) which is a responsive web tool that we have developed in order to facilitate the annotation process in a flexible, organized, efficient and tracked manner. Aspect-based SA is considered as one of the challenging tasks in SA. In this regard, an integrated lexicon and rule-based approach was employed to extract explicit and implicit aspects and their sentiment classification. This model utilized the manually generated lexicons in this dissertation with hybrid rules to handle some of the key challenges in aspect-based SA in particular and SA in general. This approach reported high performance results through an integrated lexicon and rule-based model. The approach confirmed that integrating sentiment and aspects lexicons with various rules settings that handle various challenges in SA such as handling negation, intensification, downtoners, repeated characters and special cases of negation-opinion rules outperformed the lexicon baseline and other rules combinations.
DISSERTATION WITH DISTINCTION
aspect-based sentiment analysis, app store reviews, aspect extraction, building dataset, sentiment classification, Government mobile apps