BRF: A Framework of Retrieving Brand Names of Products in Auction Sites

Abstract

Online auction sites give sellers extreme high degree of freedom to fill in the product information so that they can promote their products to attract bidders in many ways. One of the most popular ways to promote is to add brand names and model names in their product titles. However, the side effect of this promotion way is that the search results are seriously irrelevant to what users expect, especially when brand names are used as query terms. In this paper, we target at the problem of retrieving the brand name of a product from its title. First, the root causes could be categorized into three types by observing the real data on the online auction site of Yahoo! Taiwan. Then, a brand-retrieving framework BRF is proposed. Specifically, BRF first eliminates those brand and model names, which may not be the actual brand name of this product, in a product title; then BRF represents a product title by selecting representative keywords with their importance; finally, BRF models the problem as a classification problem which identify what the brand name (class) of a product title is. Extensive experiments are then conducted by using real datasets, and the experimental results showed the effectiveness of BRF. To best of our knowledge, this is the first paper to design a mechanism of retrieving the brand names of products in auction sites.