A comparison of collaborative-filtering recommendation algorithms for e-commerce

Various Collaborative Filtering (CF) recommendation algorithms characterize consumers and products by the data available about consumer-product interactions in e-commerce applications. The user-based algorithm predicts a target consumer's future transactions by aggregating the observed transactions of similar consumers. The item-based algorithm computes product similarities instead of consumer similarities and gives the products' potential scores for reach consumer. The generative-model algorithm uses latent class variables to explain the patterns of interactions between consumers and products. The spreading-activation algorithm addresses the sparsity problem by exploring transitive associations between consumers and products in a bipartite consumer-product graph. The link-analysis algorithms adapts Hypertext-Induced Topic Selection (HITS) algorithm in the recommendation context.