HINGE_code

HINGE_code copied to clipboard

Hi dingqi @dingqi, Thanks for your contributions! But I have a question about how to determine the base triplet. Do the first two entities and relation in a n-ary fact...

Finnhh

Hi, Thanks for the amazing work. I notice that in Table 2 "Original" row: the MRR values equal the Hits@1 values in the results of m-TransH and RAE. This seems...

PlusRoss

JF17K dataset is reduced

3

Paolo, Have you filtered the JF17K dataset? Why is the test dataset reduced so much?

JiyaoWei

JF17K的测试集大小为论文table1所标称的四分之一

1

您好，我在复现您的代码的时候发现了这个问题，JF17K的测试集大小是有论文里ｔａｂｌｅ1所称的四分之一，我算了一下比例是42.39，和您论文里给的42.4是一样的，想问一下这个四分之一随机取的吗？这样做的原因是因为这个的验证太慢了吗？我复现时，训练速度还好，预测很慢，ｗｉｋｉｐｅｏｐｌｅ尤其慢。对于JF17K，我把验证集加到测试集里面进行测试效果会有下降。 复现结果中，我的ｗｉｋｉｐｅｏｐｌｅ的　ｖａｌｕｅ　ｐｒｅｄｉｃｔｉｏｎ各项指标和您论文所给出入较大，我猜测可能是因为数据量太少了，其他指标都差不多，就这一项和您的差别比较大。 祝您身体健康，万事顺利，期待您的回复。

chandeler

Hello, Thank you for great work. I was wondering whether you could elaborate on the followings 1. What is the motivation of using different learning rates ? 2. What is...

Demirrr