Eric Wu

> I experimented more epochs with other models such as tucker to see that the learning approach is not right. sorry, I notice this too and I found another paper...

> Not sure what you mean by `valid`. Experimentally, the model gives top10=33% for fb15k-237 which is very low for the task. So experimentally, we can not reach to any...

Thanks for your efforts to adjust the model, I also personally think the encoding process is useful and I am currently trying to modify it for my work. Thanks very...