Accurate Product Name Recognition from User Generated Content

This paper presents the solution of the team "ISSSID" for the Consumer Products Contest #1(CPROD1) of ICDM 2012. The contest provides a dataset including hundreds of thousands of text items, a product catalog with over fifteen million products, and hundreds of manually annotated product mentions. The goal of the competition is to automatically recognize product mentions in the textual content and disambiguate which product(s) in the product catalog are referenced by the mentions. We propose a hybrid approach which combines the results obtained by several separately trained recognition models. Specifically, the approach uses a standard matching model, a rule template model, and a conditional random field model, and finally combines the results using a blending model. The proposed approach achieves the best performance in the contest.