A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network
NAACL-2018
Reference format
:
GB/T7714:
Nguyen D Q , Nguyen T D , Nguyen D Q , et al. A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network[C]// Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers). 2018.
Datasets
:
WN18RR、FB15k-237
Problems
:
ConvE focus on the local relationships among different dimensional entries in each of vh or vr.
Does not observe the global relationsships among the dimensional enties of an triple (includes vt).
Ignores the transitional characteristic in transition-based models, which is one of the most useful intutions for the task.
Score Function
:
其中Ω与W共享参数
Loss Function
:
Experiments
:
Datasets | MR | MRR | Hits@10 |
---|---|---|---|
FB15k-237 | 257 | 0.396 | 51.7 |
WN18RR | 2554 | 0.248 | 52.5 |
可参考点:
1、为什么不使用数据量更大的FB15k
与WN18
?
答:FB15k
与WN18
包含了很多可逆关系,比如<A, is_a_friend_of ,B>
,这类关系很容易逆向推出<B, is_a_friend_of, A>
,对该类关系进行剔除后生成FB15K-237
与WN18RR
。
2、简单修改Baseline的计算方法也许可以取得较好的结果。
3、不同的负采样方法存在一定的优劣性,本文采用的是Bernoulli trick
。
原文PDF与开源代码已上传