hkuds / kgrec Goto Github PK

View Code? Open in Web Editor NEW

55.0 2.0 3.0 28.45 MB

[KDD'2023] "KGRec: Knowledge Graph Self-Supervised Rationalization for Recommendation"

Home Page: https://arxiv.org/abs/2307.02759

License: Apache License 2.0

Python 93.88% C++ 6.12%

graph-neural-networks invariant-learning knowledge-graph rationalization recommender-systems self-supervised-learning

kgrec's Introduction

Introduction

This is the Pytorch implementation for our paper at KDD'23: Knowledge Graph Self-Supervised Rationalization for Recommendation.

Environment Dependencies

You can refer to requirements.txt for the experimental environment we set to use.

run KGRec

Simply use:

python run_kgrec.py --dataset [dataset_name]

And the hyperparameters we use are fixed according to the dataset in KGRec.py.

Baseline Models (KGCL, KGIN)

We also implement KGCL and include the original KGIN release in our repository. For example, to run KGCL, you may execute:

alibaba-ifashion

python run_kgcl.py --mu 0.7 --tau 0.2 --cl_weight 0.1

last-fm

python run_kgcl.py --mu 0.5 --tau 0.1 --cl_weight 0.1

mind

python run_kgcl.py --mu 0.6 --tau 0.2 --cl_weight 0.1

Citation

Please kindly cite our work if you find our paper or codes helpful.

@inproceedings{yang2023knowledge,
  title={Knowledge graph self-supervised rationalization for recommendation},
  author={Yang, Yuhao and Huang, Chao and Xia, Lianghao and Huang, Chunzhen},
  booktitle={Proceedings of the 29th ACM SIGKDD conference on knowledge discovery and data mining},
  pages={3046--3056},
  year={2023}
}

kgrec's People

Contributors

Stargazers

Watchers

Forkers

orennee liuqiangh genius-pig

kgrec's Issues

加速模型訓練

作者您好,
我照您原本設置執行run_kgrec.py
以下為各個dataset下一個epoch之執行時間
last-fm : 15min

mind-f : 2h26min

alibaba-fashion : 32min

執行環境 :
system : Ubuntu 20.04.5 LTS
CPU : Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz
GPU : NVIDIA GeForce RTX 3090

檢查GPU和CPU皆正常運作
請問該如何使執行時間如您在之前issue所述 : 800 secs for iFashion and Last-FM, and 700 secs for MIND
謝謝

ran code on the ml-10m dataset

Hello, author. I ran your code on the ml-10m dataset and found that the experiment did not work well, do you know why?

环境配置

您好，可以问下代码的大概还环境配置吗？比如python、pytorch以及用到的一些主要库的版本，，readme文件和论文里没有写。不胜感激

Dataset

作者您好,想再請教關於dataset的問題:
以last-fm為例,論文當中數據如圖

1.interaction 數量對不上
論文中interactions數量為3034796
但是train.txt 之interaction 數量為1289003
test.txt 之interaction 數量為423635

2.entity
論文中entity數量為58266
但程式碼中n_entities數量為106389

3.dataset split 不一致
論文中寫說train/valid/test = 7:1:2
程式碼當中只有train/test = 1289003:423635 約為3:1

想知道為何會有如此不一致之情況?

4.test interval和early stop step設定的原因?
test_interval = 10 if args.dataset == 'last-fm' else 1
early_stop_step = 5 if args.dataset == 'last-fm' else 10

5.test time
我在last-fm執行kgrec 程式test time 跟train time 差不多久
在alibaba-fashion 執行kgrec程式test time是train time的3倍
請問該如何加速test time(參考您之前在mind執行的log,test time是遠小於train time的)

謝謝

Does batch size affect the results?

Hello, does the batch size affect the result? I found that using the default batch size:1024 takes a long time to run, so I want to make the batch size a little bit bigger to speed up, thank you!

processing dataset

Could you explain how your dataset is processed？Thanks

Thank you very much for sharing
i run the code python run_kgrec.py --dataset last-fm to process the dataset
When I ran it(python run_kgrec.py --dataset last-fm) the second time to train
i get this error

NameError: name 'test_user_set' is not defined

experimental result

Thank you very much for sharing the code

after running the code python run_kgrec.py --dataset last-fm
Where can I view the experimental results，ndcg or recall

程式碼與論文設計不一致

作者您好,
在trace完程式碼以後,有些問題想要請教您:
1.KGRec.py combine two mask,而論文沒有此設計
2.為何在class KGRec 的forward 函式當中,
item_attn_mean_1[item_attn_mean_1 == 0.] = 1.
item_attn_mean_2[item_attn_mean_2 == 0.] = 1.
不太理解為何要把score 最低的item設為最高?
3.計算contrastive loss時,依照公式17,分母應該會產生4倍的between_sim(j分別以v,v',v''帶入公式)而不是程式碼所寫之兩倍的between_sim
4.
公式17,
應該是把所有item下之loss相加而非程式碼之取平均
5.
公式14,
對應程式碼應該使用_bi_norm_lap(adj),而程式碼卻使用_si_norm_lap(adj)是為何?

alibaba-fashion:

index 30208 is out of bounds for dimension 0 with size 30040
Traceback (most recent call last):
  File "run_kgcl.py", line 154, in <module>
    ret = evaluator.test(model, user_dict, n_params)
  File ".../KGRec-main/utils/evaluator.py", line 139, in test
    u_g_embeddings = user_gcn_emb[user_batch]
IndexError: index 30208 is out of bounds for dimension 0 with size 30040

last-fm:

index 23566 is out of bounds for dimension 0 with size 23566
Traceback (most recent call last):
  File "run_kgcl.py", line 154, in <module>
    ret = evaluator.test(model, user_dict, n_params)
  File ".../KGRec-main/utils/evaluator.py", line 152, in test
    i_g_embddings = entity_gcn_emb[item_batch]
IndexError: index 23566 is out of bounds for dimension 0 with size 23566