Reasoning Over Relations Based on Chinese Knowledge Bases

摘要: Knowledge bases are useful resource for many applications, but reasoning new relationships between entities based on them is difficult because they often lack the knowledge of relations and entities. In this paper, we introduce novel Neural Tensor Network (NTN)[1] model to reason facts Chinese bases. We represent as an average their constituting word or character vectors, which share statistical strength entities, such Open image in window . The NTN uses a tensor network replace standard neural layer, strengthen interaction two entity vectors simple efficient way. experiments, compare several other models, results show that all models’ performance can be improved when pre-trained from unsupervised large corpora don’t have advantage. outperforms others reachs high classification accuracy 91.1% 89.6% using random respectively. Therefore, segmentation task, initialization with feasible choice.