A Comparative Study of Recommendation Algorithms in E- Commerce Applications

We evaluate a wide range of recommendation algorithms on e-commerce-related datasets. These algorithms include the popular user-based and item-based correlation/similarity algorithms as well as methods designed to work with sparse transactional data. Data sparsity poses a significant challenge to recommendation approaches when applied in ecommerce applications. We experimented with approaches such as dimensionality reduction, generative models, and spreading activation, which are designed to meet this challenge. In addition, we report a new recommendation algorithm based on link analysis. Initial experimental results indicate that the link analysis-based algorithm achieved the overall best performances across several e-commerce datasets.