Whereas recommender system (RS) is ubiquitous in e-commerce platforms, recent years have seen grievous adversarial attacks on RS. However, no prior studies have evaluated RS’s adversarial vulnerability of utilizing online reviews. In this work, we follow the guidelines of adversarial robustness theory and adopt computational design science paradigm to design a novel “Min-Max” problem-based framework for assessing and enhancing adversarial robustness of review-based RS (R-RS). The framework includes an assessment component called Anchor Imitator (AIM) for crafting adversarial samples, and three enhancement components for copying with adversarial vulnerability, involving stochastic recommending process (SRP) that increases the difficulty of obtaining model information, weighted input dropout (WID) that reduces sensitivity on sensitive words, and weighted adversarial contrastive learning (WACL) that learns robust feature. We evaluate the devised framework on ground truth datasets, results demonstrate that R-RS is vulnerable to adversarial attack and the enhancement components significantly improve the adversarial robustness of R-RS.


Paper Number 1793; Track AI; Complete Paper



When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.