kNN Versus SVM in the Collaborative Filtering Framework

摘要: We present experimental results of confronting the k-Nearest Neighbor (kNN) algorithm with Support Vector Machine (SVM) in collaborative filtering framework using datasets different properties. While is usually used for tasks, considered a state-of-the-art classification algorithm. Since can also be interpreted as classification/regression task, virtually any supervised learning (such SVM) applied. Experiments were performed on two standard, publicly available and, other hand, real-life corporate dataset that does not fit profile ideal data filtering. conclude quality recommendations highly dependent data. Furthermore, we see kNN dominant over SVM standard datasets. On high level sparsity, fails it unable to form reliable neighborhoods. In this case outperforms kNN.