Mastering Machine Learning with R - Second Edition

Item-based collaborative filtering

As you might have guessed, IBCF uses the similarity between the items and not users to make a recommendation. The assumption behind this approach is that users will prefer items that are similar to other items they like (Hahsler, 2011). The model is built by calculating a pairwise similarity matrix of all the items. The popular similarity measures are Pearson correlation and cosine similarity. To reduce the size of the similarity matrix, one can specify to retain only the k-most similar items. However, limiting the size of the neighborhood may significantly reduce the accuracy, leading to poorer performance versus UCBF.

Continuing with our simplified example, if we examine the following matrix, with k=1 ...