ruizhang-ai / sign-detecting-beneficial-feature-interactions-for-recommender-systems Goto Github PK

Dear researchers,
I read your paper and it is so interesting. However, I would try to reproduce your experiments and I have on question: I run your code but just achieve the results(auc:0.8942, acc:0.8855 on movie-tags) with 300 epochs, i want to know how to achieve the experiment data in your paper.

Thank you very much,
wangzhen38

请问LinkPred类的forward函数中的torch.log2()函数和
s = torch.sigmoid(sum_log/self.temp)
s = s * (self.inter_max - self.inter_min) + self.inter_min
到底是干什么用的呢？

When I follow the "read me" in code, it turns out with "run out of input". How can I deal with it?

1. 从论文以及代码上看，分类问题也是用MSE_loss，请问我理解的对吗，如果是的话，为什么要这样用呢？

2. 边的权重f_ep，我理解通过L0近似方法得到的是 0 <= s <=1，为什么case study里有负的权重呢？

谢谢~

请问processed文件里的frappe.dataset是空的吗 它有什么作用

This is an interesting work. When I realized the baseline of Frappe data set, I found that the evaluation metrics were better than those reported in this paper. I want to know whether your baseline hyperparameter setting is consistent with the model proposed in this paper or whether grid search is the best.

1. 根据SIGN_main.py，应该将命令行中的参数predict_edge修改为pred_edges；
2. 需要在processed/frappe/目录下，新建一个空的frappe.dataset文件；