lianhaimiao / attentive-group-recommendation

Attentive Group Recommendation

Attentive Group Recommendation

This is our implementation for the paper:

Da Cao, Xiangnan He, Lianhai Miao, Yahui An, Chao Yang, and Richang Hong. 2018. Attentive Group Recommendation. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval (SIGIR '18). ACM, New York, NY, USA, 645-654.

In order to learn the group interest, we use attention mechanism to learn the aggregation strategy from data in a dynamic way.

Please cite our SIGIR'18 paper if you use our codes. Thanks!

BibTeX:

@inproceedings{Cao2018Attentive,
 author = {Cao, Da and He, Xiangnan and Miao, Lianhai and An, Yahui and Yang, Chao and Hong, Richang},
 title = {Attentive Group Recommendation},
 booktitle = {The 41st International ACM SIGIR Conference on Research \&\#38; Development in Information Retrieval},
 series = {SIGIR '18},
 year = {2018},
 isbn = {978-1-4503-5657-2},
 location = {Ann Arbor, MI, USA},
 pages = {645--654},
 numpages = {10},
 url = {http://doi.acm.org/10.1145/3209978.3209998},
 doi = {10.1145/3209978.3209998},
 acmid = {3209998},
 publisher = {ACM},
 address = {New York, NY, USA},
 keywords = {atention mechanism, cold-start problem, group recommendation, neural collaborative filtering, recommender systems},
}

Environment Settings

We use the framework pytorch.

• pytorch version: '0.3.0' or '1.x'
• python version: '3.5'

It's better to use Pytorch 1.x to run the code.

Thanks zanshuxun for providing the newest pytorch version code.

Example to run the codes.

Run AGREE:

python main.py

After training process, the value of HR and NDCG in the test dataset will be printed in command window after each optimization iteration.

Output:

AGREE at embedding size 32, run Iteration:30, NDCG and HR at 5
...
User Iteration 10 [449.8 s]: HR = 0.6216, NDCG = 0.4133, [1.0 s]
Group Iteration 10 [471.9 s]: HR = 0.5910, NDCG = 0.4005, [23.0 s]

Parameter Tuning

we put all the papameters in the config.py

Dataset

We provide one processed dataset: CAMRa2011.

Because we have another paper use the MaFengWo dataset are under reviewing, so we can't release MaFengWo dataset now.

group(user) train.rating:

• Train file.
• Each Line is a training instance: groupID(userID)\t itemID\t rating\t timestamp (if have)

test.rating:

• group(user) Test file (positive instances).
• Each Line is a testing instance: groupID(userID)\t itemID\t rating\t timestamp (if have)

test.negative

• group(user) Test file (negative instances).
• Each line corresponds to the line of test.rating, containing 100 negative samples.
• Each line is in the format: (groupID(userID),itemID)\t negativeItemID1\t negativeItemID2 ...

attentive-group-recommendation's Issues

Cloud you please tell me if the order of dataset(train.rating) is sorted by click time?

Hi,

After read AGREE paper and code, this work is great! And I want to use this dataset in my project which need the timestamp infomation. Or I want to know if the order of dataset(train.rating) is sorted by click time?

Thanks!

should I use pytorch 0.3.0 in GPU or CPU version? Or pytorch 1.0.1 can work out?

I seem to have a problem with the code. The error is "TypeError: torch.index_select received an invalid combination of arguments - got (torch.FloatTensor, int, !torch.IntTensor!), but expected (torch.FloatTensor source, int dim, torch.LongTensor index)". I wonder if I get the pytorch version wrong. Now I'm using the pytorch 0.3.0 and python 3.5.4.

Can you please provide the code of the baseline models?

Can you please provide the code of the baseline models?

您好，在运行代码时遇到了内存溢出的问题。

运行错误如下：
RuntimeError: [enforce fail at ..\c10\core\CPUAllocator.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 986880 bytes. Buy new RAM!

----》错误指向这一行 training(agree, dataset.get_group_dataloader(config.batch_size), epoch, config, 'group')，请问这个可以通过减小batch_size来处理吗？ 谢谢

problem with install torch=0.3.0

I have a problem with install torch=0.3.0. I try to install by pip and Conda. but I get these errors:

1. with pip:

pip install torch==0.3.0
Collecting torch==0.3.0
Could not find a version that satisfies the requirement torch==0.3.0 (from versions: 0.1.2, 0.1.2.post1, 0.1.2.post2)
No matching distribution found for torch==0.3.0

2. with Conda

(C:\Users\ASUS\Anaconda3) C:\Users\ASUS>conda install pytorch=0.3.0 -c pytorch
Fetching package metadata .....
CondaHTTPError: HTTP 000 CONNECTION FAILED for url https://repo.continuum.io/pkgs/main/win-64/repodata.json.bz2
Elapsed: -

An HTTP error occurred when trying to retrieve this URL.
HTTP errors are often intermittent, and a simple retry will get you on your way.
ConnectTimeout(MaxRetryError("HTTPSConnectionPool(host='repo.continuum.io', port=443): Max retries exceeded with url: /pkgs/main/win-64/repodata.json.bz2 (Caused by ConnectTimeoutError(<urllib3.connection.VerifiedHTTPSConnection object at 0x0000023DB7E31358>, 'Connection to repo.continuum.io timed out. (connect timeout=9.15)'))",),)

i have receive this issue,can you give me some advice? thanks very much!!

AGREE at embedding size 32, run Iteration:30, NDCG and HR at 5Traceback (most recent call last): File "E:/Attentive-Group-Recommendation-master/main.py", line 101, in training(agree, dataset.get_user_dataloader(config.batch_size), epoch, config, 'user') File "E:/Attentive-Group-Recommendation-master/main.py", line 62, in training print('Iteration %d, loss is [%.4f ]' % (epoch_id, np.mean(losses))) File "D:\anaconda\lib\site-packages\numpy\core\fromnumeric.py", line 3118, in mean out=out, **kwargs) File "D:\anaconda\lib\site-packages\numpy\core_methods.py", line 85, in _mean ret = ret.dtype.type(ret / rcount)AttributeError: 'torch.dtype' object has no attribute 'type'

您好，运行main.py后只能训练模型吗？如何得到推荐结果呢？

Can MaFengWo dataset be released now?

I see the unreleased reason in README.md which you wrote last year. I want to know if the dataset can be released now?

Could you tell me how to process original data to *train, *test,*test.negative data?

i want to use your mode try some other data in bioinformatics, thanks!

Could you please provide the CAMRa2011 datasets?

Hi,
It seems that you provide the processed CAMRa2011 datasets and I want to make some experiments on this datasets.
So could you please provide the original CAMRa2011 datasets?

您好，请问您是怎么进行负样本的选取的呢

Why can you achieve such high training speed?

Excuse me, After reading your thesis and code, i have benefited a lot. Again, there are some doubts, I hope you can help answer

1）Is the training speed displayed by REAME.md (shown below) based on CAMRa2011?

AGREE at embedding size 32, run Iteration:30, NDCG and HR at 5
...
User Iteration 10 [449.8 s]: HR = 0.6216, NDCG = 0.4133, [1.0 s]
Group Iteration 10 [471.9 s]: HR = 0.5910, NDCG = 0.4005, [23.0 s]

2）If so, do you run with gpu or cpu? I use your code directly and find that I can't reach this speed.

Looking forward for your response, thank you

KeyError: tensor(259)

In main.py:

def grp_forward(self, group_inputs, item_inputs):
    group_embeds = Variable(torch.Tensor())
    item_embeds_full = self.itemembeds(Variable(torch.LongTensor(item_inputs)))
    for i, j in zip(group_inputs, item_inputs):
        members = self.group_member_dict[i]   ## Showing Error here
        members_embeds = self.userembeds(Variable(torch.LongTensor(members)))

Error:

AGREE at embedding size 32, run Iteration:30, NDCG and HR at 5
E:\Anaconda3\lib\site-packages\torch\nn\functional.py:1006: UserWarning: nn.functional.sigmoid is deprecated. Use torch.sigmoid instead.
warnings.warn("nn.functional.sigmoid is deprecated. Use torch.sigmoid instead.")
Iteration 0, loss is [0.9998 ]
Traceback (most recent call last):
File "C:/Users/Leela Sravani/Desktop/RECO PAPERS/Attentive-Group-Recommendation-master/main.py", line 103, in
training(agree, dataset.get_group_dataloader(config.batch_size), epoch, config, 'group')
File "C:/Users/Leela Sravani/Desktop/RECO PAPERS/Attentive-Group-Recommendation-master/main.py", line 50, in training
pos_prediction = model(user_input, None, pos_item_input)
File "E:\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "C:\Users\Leela Sravani\Desktop\RECO PAPERS\Attentive-Group-Recommendation-master\model\agree.py", line 34, in forward
out = self.grp_forward(group_inputs, item_inputs)
File "C:\Users\Leela Sravani\Desktop\RECO PAPERS\Attentive-Group-Recommendation-master\model\agree.py", line 47, in grp_forward
members = self.group_member_dict[i]
KeyError: tensor(259)

Process finished with exit code 1

Kindly help solving this..

您好，Shared Hidden layers共享传播层的实现是用了高阶API吗，代码看不太懂

您好，Shared Hidden layers这个模块的代码我在网络传播层中没找到在哪里实现的，可以麻烦告知一下吗？谢谢！

Can you help me? I can't reproducing the preformance on your MaFengWo dataset with your code.

大佬 请问马蜂窝数据集可以公开了吗

包含真实群组信息的数据集是真少。。。求马蜂窝数据集
此外 您用马蜂窝数据集的论文发了吗 想学习一波