rootlu / sian

Code and data for ECML-PKDD paper "Social Influence Attentive Neural Network for Friend-Enhanced Recommendation"

SIAN

Code and data for ECML-PKDD paper "Social Influence Attentive Neural Network for Friend-Enhanced Recommendation"

Requirements

• Python 2.7
• PyTorch 0.4.1
• numpy
• scipy
• My machine with two GPUs (NVIDIA GTX-1080 *2) and two CPUs (Intel Xeon E5-2690 * 2)

Description

├── baselines  # baseline code
│   ├── Eval4Baselines.py

├── code  # Our Model: SIAN
│   ├── Attention.py  # attention layer
│   ├── DataUtil.py  # data loader 
│   ├── Evaluation.py  # model evaluation
│   ├── FeatureAgg.py  # attentive feature aggregator 
│   ├── Fusion.py  # feature fusion layer 
│   ├── HeteInf.py  # the main class for SIAN
│   ├── InfluenceProp.py  # social influence coupler 
│   ├── Logging.py  #log

│   └── trainHeteInf.py  # the main function for SIAN
└── data  # dataset
    ├── Data4Baselines.ipynb  #
    ├── DataProcessor.ipynb
    ├── ItemProfileEmbed.ipynb
    ├── WechatTencent.ipynb
    ├── wxt  # FWD dataset 
    │   ├── biz2id
    │   ├── biz_profile.npy
    │   ├── item2id
    │   ├── item_profile.npy
    │   ├── user2id
    │   ├── user_profile.npy
    │   ├── wxt.att.analysis
    │   ├── wxt.interaction.graph
    │   ├── wxt.item.biz
    │   ├── wxt.social.graph
    │   ├── wxt.test.rating.712
    │   ├── wxt.train.rating.712
    │   ├── wxt.user.biz
    │   └── wxt.val.rating.712
    ├── wxt.ipynb
    ├── yelp  # yelp dataset
    │   ├── item_profile.npy
    │   ├── user_profile.npy
    │   ├── yelp.att.analysis
    │   ├── yelp.interaction.graph
    │   ├── yelp.social.graph
    │   ├── yelp.test.rating.712
    │   ├── yelp.train.rating.712
    │   └── yelp.val.rating.712
    ├── yelp.ipynb


├── log  # saved log file
│   ├── wxt.0.0.6023.0.35225.model

Dataset

FWD dataset (i.e., wxt data) can be downloaded from Google Drive and BaiduYun （提取码：i6qy）

Reproducing results in the paper

Load the saved models in log/ dir.

Training

python trainHeteInf.py  --help

Reference

@inproceedings{Yuanfu2020SIAN,
  title={Social Influence Attentive Neural Network for Friend-Enhanced Recommendation},
  author={Yuanfu Lu, Ruobing Xie, Chuan Shi, Yuan Fang, Wei Zhang, Xu Zhang, Leyu Lin.}
  booktitle={Proceedings of ECML-PKDD},
  year={2020}
}

sian's Issues

您好，可不可以发布一下豆瓣数据集，想学习一下

或者发一下邮箱可以吗？[email protected]

Error in Using wxt dataset

2020.07.19-15:25:05
Namespace(batch_size=1024, biz_profile='biz_profile.npy', dataset_name='wxt', emb_size=64, epochs=100, eval_num=100, inf_flag=True, item_num=7900, item_profile='item_profile.npy', lr=0.0001, on_YARD=False, profile_size=150, save_model=True, seed=72, test_batch_size=1024, top_k=10, use_cuda=False, user_num=8163, user_profile='user_profile.npy', weight_decay=0.0005, worker_num=0)
Using cuda...
Loading data...
--- # interaction links: 2465675, social links: 8639884.0, #u_bizs: 71983, i_bizs:22218
--- #train size:1725973 #user:25049, #item:22218
--- #valid size:246568 #user:13424, #item:8075
--- #test size:493134 #user:16659, #item:11511
Load data finished 19.68671464920044s
Initializing model...
Traceback (most recent call last):
File "./SIAN/code/trainHeteInf.py", line 486, in
main(args)
File "./SIAN/code/trainHeteInf.py", line 271, in main
item_num=args.item_num, biz_num=args.biz_num,
AttributeError: 'Namespace' object has no attribute 'biz_num'

May I know what the correct value of biz_num is?

Will Douban dataset be released?

I have run experiments on Yelp dataset and found the result in the paper is reproducible. I wonder if the Douban dataset could be made public so that a more comprehensive analysis can be done on the proposed model.

请问yelp数据集里的act列是什么id呢？

看代码里有 user_id \t item_id \t label \t act_list 这样的注释，act_list 表示的是 user 的所有好友吗？
注意到同一个user_id，其act_list 并不等长，但同一user的多个样本中，短act_list 总是长act_list 的子集，这里是为什么吗？是因为数据集处理是根据rating时间以及建立好友时间来做的，rating之后建立的好友关系并未放入模型训练是吗？

可不可以给一下处理后的doubun数据集，在学习研究您的论文和代码，感恩