Research in the International Journal of Information and Communication Technology discusses the development of an artificial intelligence (AI) system that combines text, images and reviewer behaviour to detect and trace fake e-commerce reviews. The system could address the growing challenge faced by online marketplaces as deceptive feedback becomes increasingly sophisticated.
The team used a multimodal approach to analyse several types of data at once rather than relying solely on an examination of written comments. Existing systems often focus on review text or simple behavioural indicators, making them vulnerable to fabricated reviews paired with misleading images.
To improve detection, the researchers used a text convolutional neural network. This is a machine-learning model designed to identify patterns in language. In parallel, a pre-trained language model was employed that captures broader semantic meaning. The team adds that information about reviewers was also incorporated into the analysis as well as images attached to reviews. The images were analysed using a residual network, a deep-learning architecture used in computer vision.
The system then brings together these various signals to work out whether a particular review is genuine or not. A Transformer model, widely used in modern AI systems, could then be used to trace the origins and spread of a review flagged as suspicious. Tests on large-scale datasets showed measurable gains over existing methods, the team reports.
Duan, B. (2026) ‘Precise identification and traceability of fake e-commerce reviews integrating multimodal semantic understanding’, Int. J. Information and Communication Technology, Vol. 27, No. 35, pp.81–102.
No comments:
Post a Comment