CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Constrained matrix completion algorithm considering individual differences Authors:  Yuki Morioka - Doshisha University (Japan) [presenting]
Kensuke Tanioka - Wakayama Medical University (Japan)
Hiroshi Yadohisa - Doshisha University (Japan)
Abstract: The matrix completion problem has attracted considerable attention for recommendation systems, largely through the famous Netflix competition. Recently, many matrix completion methods have been proposed and evaluated in terms of estimation accuracy or calculation speed. Various matrix completion methods for recommendation systems are applied to data where the row, column, and value indicate the user, item, and evaluation, respectively. However, the existing methods have a limitation in that the estimation accuracy is low for data where individual user evaluations tend to be lower or higher because these methods do not consider individual differences among users. Moreover, if the data consists of different scales, such as a nominal scale or ordered scale, it is difficult to evaluate the information and estimate values correctly. Therefore, we propose a novel matrix completion method that considers users individual differences and mixed scales of user's external information as dummy variables. The proposed method estimates each parameter using biased inductive matrix completion. We evaluate the estimation accuracy by conducting numerical experiments including a simulation study and real data analysis.