A Multimodal AI System Concept for Large‑Scale Consumer Opinion Analysis and Dual‑Quality Detection Across European Markets

Abstract

This paper presents the concept of a multimodal artificial intelligence system designed for large-scale acquisition, integration, and analysis of consumer opinions to support the detection of the so-called dual quality phenomenon in products available on European markets. The proposed approach combines multilingual natural language processing, image analysis, automatic speech recognition, as well as product identification and entity-linking mechanisms.The system covers heterogeneous data sources, including online stores, marketplaces, blogs, forums, social media platforms, and audio-video materials. It uses dedicated data acquisition components, structured data standards, and methods based on large language models for processing unstructured content. A key stage of the pipeline is the transformation of source data into source-product-opinion relations, which are subsequently subjected to anonymization, deduplication, credibility assessment, sentiment analysis, and detection of signals potentially related to dual quality.An essential element of the concept is the aggregation of products originating from diverse sources and markets, despite inconsistent names, missing identifiers, and language differences. This process combines automatic methods with expert validation in a human-in-the-loop model, enabling the correction of product assignments, interpretation of ambiguous content, and adjustment of the scope of analysis to the objective of a given study.The proposed concept addresses the needs of institutions responsible for market surveillance by supporting the prioritization of regulatory actions and the selection of products requiring further analysis, including laboratory testing. The system is not intended to unequivocally confirm the occurrence of dual quality, but rather to indicate cases in which multilingual and multimodal consumer data provide grounds for in-depth verification.

Keywords: dual quality detection, consumer opinion mining, multimodal data processing, product entity linking, AI in product data quality, sentiment analysis, data aggregation, market surveillance

DOI: 10.54941/ahfe1008169