- Title
- Machine intelligence for analysing and understanding online customer reviews
- Creator
- Satiabudhi, Gregorius
- Relation
- University of Newcastle Research Higher Degree Thesis
- Resource Type
- thesis
- Date
- 2022
- Description
- Research Doctorate - Doctor of Philosophy (PhD)
- Description
- Online reviews are increasingly important for decision making. These days, customers read online reviews to obtain the opinions of other customers to understand their sentiments about a product before making a purchase. Marketers also acknowledge the importance of online reviews by analysing the opinions of customers on their products. This feedback can be used to understand the trends of the masses, and evaluate brand perception and customer satisfaction levels, and this information can help improve product success and drive sales. However, reading all the online reviews of products or services is challenging, mainly because of the massive quantity of reviews found on various review websites and social media. Thus, an automated system capable of analysing reviews is required. When a system that allows automated review analysis is in place, potential buyers can obtain a holistic view of a product rapidly and accurately. Similarly, marketers would be able to summarise the opinions of buyers of their products. However, the financial effects of positive reviews have prompted some fraudulent sellers to generate fake product reviews to either promote their products or discredit competing products. Many sellers or vendors engage in fraudulent behaviours to boost the rating of their products by colluding with reviewers to provide positive ratings and reviews of their products, while providing negative ratings and reviews to their competitors’ products. This fraudulent behaviour extends to the level where some companies sell their services of providing good reviews and ratings to sellers or vendors. This malicious practice can also be used to attack competitors by damaging the reputation of their products. There is a good chance for fake reviews to distort the evaluation of a product, erode trust in consumer reviews and eventually undermine the effectiveness of online markets. Unless such reviews are detected and acted upon, social media will be increasingly flooded with lies and deception, and eventually become useless from an e-commerce perspective. Automated review analysis generally involves training machines to capture discriminative features from these online reviews and accurately predict the reviewers’ sentiments about the reviewed product. Machine learning and deep learning algorithms are multi-purpose tools for classification and prediction problems. This thesis proposes a novel framework for gauging the sentiments of online reviews using several machine learning and deep learning classifiers. With this framework, the combination of proper text preprocessing and feature extraction methods, in conjunction with the chosen classifier, is also investigated. Further, three novel nature-inspired algorithms based on the particle swarm optimisation method and parallelism techniques are proposed to improve the performance of single and ensemble classifier models for sentiment polarity prediction. For fake review detection, two different approaches for input feature extraction are proposed. The first is a textual-based featuring approach that extracts features directly from text reviews. The second approach is a hybrid content-based and behaviour-based approach. Here, the research focuses on the combination of textual content and user behaviour features. Then, using the aforementioned framework, the best machine learning or deep learning model is selected. Combined with the feature extraction approaches above, the best classifier will be used to detect fake reviews. To improve the performance of fake review detection, a novel ensemble model—the multi-type classifier ensemble (MtCE)—is also proposed. Unlike other ensembles, which use only one type of single classifier, the MtCE uses several different types of machine learning and deep learning single classifiers as its base classifiers. Overall, this research provides an enhanced understanding of online customer reviews and makes them more reliable and useful by tackling the fake review issue. With more reliable and useful reviews, potential customers can better understand other buyers’ opinions of the relevant products or services and make appropriate purchase decisions. For marketers, this accurate and timely knowledge of customer opinions about their products or services can be used to make appropriate business decisions swiftly.
- Subject
- machine intelligence; sentiment analysis; online customer reviews
- Identifier
- http://hdl.handle.net/1959.13/1508185
- Identifier
- uon:56102
- Rights
- Copyright 2022 Gregorius Satiabudhi
- Language
- eng
- Full Text
- Hits: 16
- Visitors: 24
- Downloads: 10
Thumbnail | File | Description | Size | Format | |||
---|---|---|---|---|---|---|---|
View Details Download | ATTACHMENT01 | Thesis | 7 MB | Adobe Acrobat PDF | View Details Download | ||
View Details Download | ATTACHMENT02 | Abstract | 569 KB | Adobe Acrobat PDF | View Details Download |