PRISM – Predictive Review Integrity Scoring via Multimodality

Date of Award

5-9-2026

Degree Name

M.S. in Computer Science

Department

Department of Computer Science

Advisor/Chair

Tam Nguyen

Abstract

The rapid growth of Generative AI is changing how user-generated content appears on e-commerce platforms, raising new concerns about authenticity. Motivated by financial incentives and the desire to build credibility, some users now rely on AI tools to enhance or “polish” their reviews. This can include using large language models to make experiences sound more enthusiastic or persuasive, as well as generative image tools to alter visual evidence. As a result, these AI-enhanced reviews introduce a new form of misleading content that weakens consumer trust. Most existing fake review detection methods focus primarily on text analysis. However, e-commerce reviews often depend heavily on images as supporting evidence, making text-only approaches insufficient. To address this limitation, this thesis proposes PRISM (Predictive Review Integrity Scoring via Multimodality), a classification framework designed to distinguish genuine human reviews from AI-polished ones. PRISM combines textual and visual feature extraction to evaluate both writing patterns and image authenticity. By analyzing inconsistencies introduced during the AI polishing process, the proposed multimodal approach provides a more reliable method for detecting inauthentic reviews and improving trust in e-commerce marketplaces.

Keywords

Artificial Intelligence, Computer Science

Comments

OCLC No. 1591830010

Rights Statement

Copyright 2026, author.

Link to Full Text

Share

COinS
 
 
 

Links