Missing Value Analysis in User Modeling

In this paper, we address the problem of missing values in input datasets of user modeling algorithms. The origin of these missing values is not a typical one as known in general statistics, but a combination of a so called "cold start problem" and the fact that one user typically gives ratings to only a relatively small number of content items. The ratio of missing values can be extremely high, even as much as 98%. In this context the following question arises naturally "Can we apply the known missing value analysis procedures in this context". We present the experimental setup and experimental results of our approach. We also discuss known user modeling techniques together with proposed missing value management procedures. We use standard efficiency evaluation measures such as F-measure to evaluate the proposed algorithms.