Repozytorium publikacji - Politechnika Gdańska

In recent times, the number of spam reviews through various online platforms has emerged as a prime challenge, profoundly impacting businesses and consumers. These fake reviews not only distort clients’ perceptions of products and services but also erode trust within the digital ecosystem. Despite the advent of machine learning (ML) techniques for identifying spam reviews, comparing text, and pinpointing groups of spammers, there remain notable gaps in both accuracy and the combination of interactive visualization for real-time decision-making. This paper presents SpamVis, a visual interactive system that leverages deep learning (DL) and ML blended with advanced visualization techniques for spam review detection, enabling analysts to conduct complicated analytical queries. The system allows users to input via click-on or touch to generate interactive charts and plots tailored for spam review analysis. The findings of the baseline test carried out on 67,395 review texts demonstrate that Bidirectional Encoder Representations from Transformers (BERT) carried out the best accuracy (86%) compared to other models. Our outcomes suggest that SpamVis can alleviate the gaps concerning accuracy and visualization needs in contemporary techniques, guiding analysts to make informed decisions for mitigating spam reviews and enhancing consumer trust. Furthermore, SpamVis empowers users to seamlessly discover the online reviews of various social media platforms in real-time, such as Facebook, Youtube, etc., giving them practical insights to navigate the online marketplace effectively

Autorzy

• Nguyen Thanh Thao Lam,
• Nu Uyen Phuong Le,
• Md Rafiqul Islam,
• Cesar Sanin,
• prof. dr hab. inż. Edward Szczerbicki link otwiera się w nowej karcie