ASR-enhanced Multimodal Representation Learning for Cross-Domain Product Retrieval
This addresses the challenge of noisy ASR text in multimodal representation learning for e-commerce product retrieval, but it is incremental as it builds on existing multimodal and LLM methods.
The paper tackled the problem of cross-domain product retrieval in e-commerce by proposing ASR-enhanced Multimodal Product Representation Learning (AMPere), which uses an LLM-based summarizer to de-noise ASR text and combines it with visual data in a multi-branch network, resulting in improved retrieval performance as verified on a large-scale tri-domain dataset.
E-commerce is increasingly multimedia-enriched, with products exhibited in a broad-domain manner as images, short videos, or live stream promotions. A unified and vectorized cross-domain production representation is essential. Due to large intra-product variance and high inter-product similarity in the broad-domain scenario, a visual-only representation is inadequate. While Automatic Speech Recognition (ASR) text derived from the short or live-stream videos is readily accessible, how to de-noise the excessively noisy text for multimodal representation learning is mostly untouched. We propose ASR-enhanced Multimodal Product Representation Learning (AMPere). In order to extract product-specific information from the raw ASR text, AMPere uses an easy-to-implement LLM-based ASR text summarizer. The LLM-summarized text, together with visual data, is then fed into a multi-branch network to generate compact multimodal embeddings. Extensive experiments on a large-scale tri-domain dataset verify the effectiveness of AMPere in obtaining a unified multimodal product representation that clearly improves cross-domain product retrieval.