dc.description.abstract |
Customer reviews play a vital role in e-commerce. Most online customers depend on online reviews before making a purchasing decision, and the credibility of online reviews significantly impacts the company’s reputation. Given the significance of reviews in generating revenue, some business owners reward reviewers who are dishonest. Consumer reviews are continuously produced at a high volume, velocity, and degree of unstructured. Hence, they can be treated as big data and need big data analysis methods to detect deceptive reviews. Due to the challenges highlighted among different classification processes, there is a research challenge in designing an effective deceptive review detection mechanism without focusing on labelled datasets in real-time. This research combines review-centric and reviewer-centric features in the feature selection stage and finds the deceptive level of each review based on those features without focusing on classification. The purpose of this research is to use multi-agent technology, a modern trend in Artificial Intelligence, to support the deceptive review detection process to automate complex tasks such as real-time data acquisition, feature selection, and calculating deceptive levels by ensuring high accuracy. The system follows a module architecture where all modules are incorporated with agents performing module tasks by communicating, coordinating, and negotiating with each other. The fuzzy agents in the credibility analysis module incorporate review content honesty, content quality, reviewer expertise, and reviewer trustworthiness for analysing the review credibility and reviewer credibility. The deceptive levels were calculated based on the credibility values. The human evaluated results were used to compare the results of the proposed model, k-means cluster results, and outlier-based deceptive identification method. The evaluation results indicated that the accuracy of detecting deceptive consumer reviews using multi-agent technology in big data analytics is 81% where the cluster model showed 73% and outlier-based model showed 63%. Also, the evaluation showed the importance of considering both review credibility and reviewer credibility when deciding on deceptive level. Therefore, the challenges encountered in existing deceptive review detection methodologies, such as scarcity of labelled data for model training, real-time data analysis, and uncertainty of credibility, were resolved by incorporating multi-agent technology in big data analytics. The ultimate goal of noticing the misleading level of reviews is to create an assured customer who will boost the business’s revenue by expanding purchases because of the trustful and reliable reviews. |
en_US |