Large-scale Multi-class and Hierarchical Product Categorization for an E-commerce Giant

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 525–535, Osaka, Japan, December 11-17 2016. Large-scale Multi-class and Hierarchical Product Categorization for an E-commerce Giant Ali Cevahir RIT Tokyo, Rakuten Inc., Rakuten Crimson House, 1-14-1 Tamagawa, Setagaya-ku, Tokyo ali.cevahir@rakuten.com Koji Murakami RIT NY, Rakuten USA, 215 Park Avenue South, New York NY 10003 koji.murakami@rakuten.com Abstract In order to organize the large number of products listed in e-commerce sites, each product is usually assigned to one of the multi-level categories in the taxonomy tree. It is a time-consuming and difficult task for merchants to select proper categories within thousands of options for the products they sell. In this work, we propose an automatic classification tool to predict the match- ing category for a given product title and description. We used a combination of two different neural models, i.e., deep belief nets and deep autoencoders, for both titles and descriptions. We implemented a selective reconstruction approach for the input layer during the training of the deep neural networks, in order to scale-out for large-sized sparse feature vectors. GPUs are uti- lized in order to train neural networks in a reasonable time. We have trained our models for around 150 million products with a taxonomy tree with at most 5 levels that contains 28,338 leaf categories. Tests with millions of products show that our first predictions matches 81% of merchants’ assignments, when “others” categories are excluded. 1 Introduction E-commerce has grown rapidly in recent years. Giant e-commerce companies like Amazon, e-Bay, Taobao and Rakuten list millions of products on their sites sold by thousand of merchants. As of May 2016, Japan’s largest e-commerce site Rakuten Ichiba1 hosted 186 million active products sold by 43,363 different merchants. In order to organize products so that customers can navigate and search them easily, products are categorized into multi-level categories. “Women’s Fashion > Tops > Sweaters > Long- sleeved knit > Crew neck” is an example for such categories. Rakuten Ichiba contains around 30 thou- sand categories of up to 5 levels. Merchants need to manually assign each product to one of those categories, which is a tedious task and prone to error. Moreover, merchants may not be accurate while assigning products, the cateogry assignment of the same product listed by different merchants may not be consistent. Automatic category recommendation for given product information helps to solving these problems. Product classification is a text classification with a large hierarchical product taxonomy, and a lot of re- search have been conducted with various methodologies (Gupta et al., 2016; Shen et al., 2012; Kozareva, 2015; Qiu et al., 2011). Product classification includes the following challenges: 1) the products sparsely distributed in a large number of categories and data distribution is quite skewed, 2) the length of product titles and descriptions is diverse, 3) even though there are tons of product data, it is not guaranteed that pairs of current product title and assigned category are correct. In this paper we propose a large-scale classification …

Content truncated to 500 words