A comparison of several algorithms for collaborative filtering in startup stage

Collaborative filtering is becoming a popular technique for reducing information overload. Many algorithms have been proposed for collaborative filtering. The performance of a recommended system during the startup stage is crucial to the system. If recommendation is close to what an user really want, the user would be glad to use the system later, else he may never make use of it again. In this paper, we compare the performance results of four collaborative filtering algorithms applied in the startup stage of recommendation. We evaluate these algorithms using three publicly available datasets. Our experiments results show that Pearson and STIN1 methods perform better than latent class model (LCM) and singular value decomposition (SVD) methods during the startup stage. The experimental results confirm that the characteristics of datasets keep being an important factor in the performance of methods.