hkuds / dccf

[SIGIR'2023] "DCCF: Disentangled Contrastive Collaborative Filtering"

Hi there 👋

✨Welcome to the Data Intelligence Lab @ HKU!✨

🚀 Our Lab is Passionately Dedicated to Exploring the Forefront of the Data Science & AI 👨‍💻

       

dccf's Issues

关于DGCF,DGCL模型的复现

作者您好！请问一下在4.2节性能比较中，模型DGCF和DGCL是怎么复现的呢？是否有代码可以参考

model problem

when i run python DCCF_PyTorch.py --dataset gowalla --epoch 150
get this error
file"DCCF_PyTorch.py",line 53, in
data_generator = Data(args)
file"DCCF-main\utility\load_data.py" line19, in init
train_mat = pickle.load(f)

ModuleNotFoundError:No module named 'scipy.sparse._coo'

i have installed the pack in your readme.md

求教模型分析中 公式21 22 的理解

非常荣幸能够拜读您的文章，文章巧妙的将理论假设，数学推导和代码实现结合，并用实验证明的idea的有效性。
在详细阅读文章的模型分析部分时，遇到一些疑惑，还请不吝赐教：

1. 代码中可以看到用户项目意图矩阵是一个32 x 128的矩阵，公式21为对某个用户意图的the disentangled global collaborative view进行求导，想请问这里矩阵是不是一个K=32,d=128的矩阵？论文描述里面是否混用了参数d？
2. 请问公式22中的m和n是代表不同的用户吗，还是其他含义？
3. 请问古罗马I是一个判别函数？
4. 感觉公式22同时使用了复合函数求导和函数乘积的求导法则，但并没有看懂，请问可否给出详细推导过程？

祝好！

Question regarding the reported performance

Hi,
I was impressed in your previous work, LightGCL and DCCF and noticed these papers are from the same lab!
Thanks for sharing interesting works.
While reading the paper, I found that the performance you have reported in the DCCF paper is quite different from the one you have reported in LightGCL.

In LightGCL, the performance for gowalla is reported as 0.15 and the code you have uploaded achieves around 0.21.
However, in DCCF, the performance of LightGCL is reported as 0.18.
Could you give me an explanation regarding where these differences came from?

Thanks!

Questions about the data

Hi, what's the meaning of the data in the train.pkl and train_index.pkl? If I have some data like (users, items, ratings). Can I directly convert them into the format needed for the DCCF model?
Looking forward to your reply!

original data of tmall

Thank you very much for your work. Could you provide a link to the original data of tmall?

Request for datasets

Hi author, is it possible to provide the mapping file between the dataset you used and the original dataset?

On the memory consumption of DCCF training on ml20m

Hi, thank you very much for this wonderful work. I want to try to train on ml20m dataset with DCCF on RTX A5000 GPU (24GB), but it shows that there is not enough memory, is there any suggestion to solve it other than changing to a bigger GPU?

Problem about paper

Please ask if there are any problems with the redline section, especially in terms of dimensionality.

公式

请问论文中公式（1）如何理解呢，不太懂P(y,cu,cv|u,v)表示什么意思,以及公式(2)的推导？谢谢

作者您好，打扰您一下，请问能更新一下数据集么？

非常感谢您们能够分享您们的研究成果，但是我尝试运行您们提供的源码时，出现了以下问题，运行环境是按照您们的readme进行配置的，能麻烦提供一下帮助么？
train_mat = pickle.load(f)
ModuleNotFoundError: No module named 'scipy.sparse._coo'

您好！关于MAD计算

想问问有没有MAD计算的具体实现代码？

model problem

您好，在跑这篇论文模型时，遇到OSError: /home/XXXX/.conda/envs/py381/lib/python3.8/site-packages/torch_sparse/_diag_cuda.so: undefined symbol: _ZN3c106detail19maybe_wrap_dim_slowEllb问题，可以说一下怎么解决吗，万分感谢

关于4.4.1 数据稀疏性方面的性能

请问是否有具体实现代码可以参考！万分感激