Abstract
Attribute value extraction from product profiles is essential for many applications such as product retrieval, comparison, and recommendation. While existing techniques focus mainly on the extraction task, none of them deals with the problem of correcting wrong attribute values. In this paper we propose CAVE, a novel system for attribute correction and enrichment using the Question Answering (QA) paradigm. CAVE learns information from both titles and attribute tables, using encoder and language models to correct attribute values. It also has the capability to enrich existing product descriptions with new attribute values extracted from titles. To the best of our knowledge, CAVE is the first system that allows users to experiment with a number of powerful QA models and compare their performances on attribute values correction using real-word datasets.