Large-scale Item Categorization for e-Commerce

CIKM 2012:596-604
Large-scale Item Categorization for e-Commerce
Dan Shen, Jean-David Ruvini, Badrul Sarwar

This paper studies the problem of leveraging computationally intensive classification algorithms for large scale text categorization problems. We propose a hierarchical approach which decomposes the classification problem into a coarse level task and a fine level task.

A simple yet scalable classifier is applied to perform the coarse level classification while a more sophisticated model is used to separate classes at the fine level. However, instead of relying on a human-defined hierarchy to decompose the problem, we use a graph algorithm to discover automatically groups of highly similar classes.

As an illustrative example, we apply our approach to real-world industrial data from eBay, a major e-commerce site where the goal is to classify live items into a large taxonomy of categories.

In such industrial setting, classification is very challenging due to the number of classes, the amount of training data, the size of the feature space and the real world requirements on the response time. We demonstrate through extensive experimental evaluation that (1) the proposed hierarchical approach is superior to flat models, and (2) the data-driven extraction of latent groups works significantly better than the existing human-defined hierarchy.

Another publication from the same category: Machine Learning and Data Science

IEEE Computing Conference 2018, London, UK

Regularization of the Kernel Matrix via Covariance Matrix Shrinkage Estimation

The kernel trick concept, formulated as an inner product in a feature space, facilitates powerful extensions to many well-known algorithms. While the kernel matrix involves inner products in the feature space, the sample covariance matrix of the data requires outer products. Therefore, their spectral properties are tightly connected. This allows us to examine the kernel matrix through the sample covariance matrix in the feature space and vice versa. The use of kernels often involves a large number of features, compared to the number of observations. In this scenario, the sample covariance matrix is not well-conditioned nor is it necessarily invertible, mandating a solution to the problem of estimating high-dimensional covariance matrices under small sample size conditions. We tackle this problem through the use of a shrinkage estimator that offers a compromise between the sample covariance matrix and a well-conditioned matrix (also known as the "target") with the aim of minimizing the mean-squared error (MSE). We propose a distribution-free kernel matrix regularization approach that is tuned directly from the kernel matrix, avoiding the need to address the feature space explicitly. Numerical simulations demonstrate that the proposed regularization is effective in classification tasks.