| （英） |
In this paper, we verify the effectiveness of feature integration with multi-user behavior information for interest level estimation in contents. Users' behavior information during viewing contents is correlated with the users' preferences for the contents, and is effective for interest level estimation. Furthermore, since behavior information of the target user, which is the target of interest level estimation, is correlated with behavior information of other users, the accuracy of interest level estimation is expected to be improved by simultaneously using the behavior information of target user and other users. Therefore, in this paper, we focus on semi-supervised ordinally multi-modal Gaussian process latent variable model (semi-OMGP), which is one of the methods that can use behavior information of multiple users. semi-OMGP integrates features calculated from content and multi-user behavior information, and we verify the accuracy of interest level estimation based on the integrated features to confirm the effectiveness of using multi-user behavior information. |