Research and Implementation of a Brand Normalization Method across E-Commerce Platforms
Lan Yao, Zhuang Li, Tiezheng Nie, Zhibin Zhao · 2018
A brand is the basic attribute of a commodity, and it is the primary metadata for commodity information analysis. However, in different E-commerce platforms, the same brand entity may be named as different labels, which brings challenges to further data analysis on spoken language analysis, personalized recommendation and other works. Brand Entity Normalization becomes a research issue to recognize labels under a certain brand. The former methods focus on dictionary based text matching. However, it is unpractical to define matching rules in an effective way. This paper proposes a brand entity normalization algorithm integrating text similarity and commodity object similarity. Firstly, the text similarity of brand labels is computed to generate candidate brand entity set. Furthermore, the commodity information is introduced and candidate brand entities are merged by Gaussian mixture clustering. The performance of the proposed method is evaluated by involving real data from popular E-commerce platforms. Experimental results show that this method achieves solid performance in brand entity normalization.