Optimizing Multi-Relational Factorization Models for Multiple Target Relations

Multi-matrix factorization models provide a scalable and ef-fective approach for multi-relational learning tasks such as link prediction, Linked Open Data (LOD) mining, recom-mender systems and social network analysis. Such models are learned by optimizing the sum of the losses on all rela-tions in the data. Early models address the problem where there is only one target relation for which predictions should be made. More recent models address the multi-target vari-ant of the problem and use the same set of parameters to make predictions for all target relations. In this paper, we argue that a model optimized for each target relation indi-vidually has better predictive performance than models op-timized for a compromise on the performance on all target relations. We introduce specific parameters for each tar-get but, instead of learning them independently from each other, we couple them through a set of shared auxiliary parameters, which has a regularizing effect on the target specific ones. Experiments on large Web datasets derived from DBpedia, Wikipedia and BlogCatalog show the per-formance improvement obtained by using target specific pa-rameters and that our approach outperforms competitive state-of-the-art methods while being able to scale gracefully to big data