Product attribute-value extraction is a core step in building high-quality e-commerce catalogs. The task involves identifying attribute names and their corresponding values from product listings.
A new paper on arXiv notes that existing systems typically rely on either generative or tagging strategies to perform this extraction. The paper, titled "Hybrid Generative-Tagging Decoding for Efficient Product Attribute-Value Extraction," proposes a hybrid decoding approach that combines both strategies.
The abstract does not include details on how the hybrid method is implemented or how it performs relative to existing approaches. The paper is available on arXiv.