TransA 论文笔记

- Abstract

translation-based methods also suffer from the oversimplified loss metric, and are not competitive enough to model vari- ous and complex entities/relations in knowledge bases.

- Background

目前的的翻译模型是使用球形等势线，距离越近，可能性越大。在复杂关系中，是不够灵活的。

due to the inflexibility of loss metric, cur- rent translation-based methods apply spherical equipoten- tial hyper-surfaces with different plausibilities, where more near to the centre, more plausible the triple is. As illustrated in Fig.1, spherical equipotential hyper-surfaces are applied in (a), so it is difficult to identify the matched tail entities from the unmatched ones.

某种关系可能只被某几个特定纬度影响，其他的不相干维度将会变成噪声。