Product Matching with Multimodal Integration for E-Commerce Price Intelligence

Abstract

Product matching is a fundamental challenge in e-commerce, where the task is to determine whether two product listings from different sources refer to the same real- world item. Accurate identification is essential for price monitoring, stock management, and competitive intelligence, yet traditional approaches based on textual similarity or structured identifiers often struggle with incomplete and heterogeneous data. This paper presents RePrice, a hybrid product matching system that integrates text, attribute, and image similarity within a unified decision pipeline. The approach combines deterministic rules with AI-driven scoring, leveraging semantic embeddings and CLIP-based image features. Experiments on 29,000 product listings from 23 platforms demonstrate that multimodal integration significantly improves performance, achieving an F1-score of 0.75 compared to 0.42 for a text-only baseline. We analyze the contribution of each modality, provide category-level results with error analysis, and discuss the scalability and robustness of the system for real-world e-commerce applications.