SIGIR2021_Conure
One Person, One Model, One World: Learning Continual User Representation without Forgetting
Posts: https://zhuanlan.zhihu.com/p/437671278
@inproceedings{yuan2021one,
title={One person, one model, one world: Learning continual user representation without forgetting},
author={Yuan, Fajie and Zhang, Guoxiao and Karatzoglou, Alexandros and Jose, Joemon and Kong, Beibei and Li, Yudong},
booktitle={Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval},
pages={696--705},
year={2021}
}
If you want to use Conure in real production systems with hundreds of million items. I strongly suggest: (1) understand our code released here ; (2)using TFRecord (tf.data.Dataset) and tf.estimator to replace feed_dict (slow), which is several times faster (see https://zhuanlan.zhihu.com/p/53345706); (3) contact [email protected] if you could not achieve expected results. (E.g., No personalization for new user recommendation);(4) For simplicity, we use standard dense matrix to store the binary mask matrix. You can replace it with the sparse matrix which is much more parameter efficient; (5) Please also note that the code attached here was rewritten by Fajie after leaving Tencent. Though it was not the original code used for this paper, you should be able to reproduce all results reported in the paper.
NextItNet pytorch version: https://github.com/syiswell/NextItNet-Pytorch
conure_tp_t1.py: Conure is trained on Task1, and once converged it will be pruned.
conure_ret_t1.py: Conure retrains the pruned architecture of Task1
conure_tp_t2.py: Conure is trained on Task2 and once converged it will be pruned.
conure_ret_t2.py: Conure retrains the pruned architecture of Task2
Large-scale Recommendation Dataset for pretraining，transfer learning (crosss-domain recommendation) and user representation learning：
Download the TTL dataset from: https://drive.google.com/file/d/1imhHUsivh6oMEtEW-RwVc4OsDqn-xOaP/view?usp=sharin or the ML dataset from: https://drive.google.com/file/d/1-_KmnZFaOdH11keLYVcgkf-kW_BaM266/view?usp=sharing
TTL Dataset: 可用于推荐系统预训练，迁移学习，跨域推荐，冷启动推荐，用户表征学习，自监督学习等任务。
Running our code:
Put these dataset on Data/Session
FOllowing these steps:
python conure_tp_t1.py After convergence (it takes more than 24 hours for training). We suggest 4 iterations. Parameters will be automatioally saved.
python conure_ret_t1.py You can manually stop this job if the results are satisfied (better than results reported in conure_tp_t1.py). Parameters will be automatioally saved.
python conure_tp_t2.py
python conure_ret_t2.py
python conure_tp_t3.py
python conure_ret_t3.py
python conure_tp_t4.py
python conure_ret_t4.py
Environments
- Tensorflow (version: 1.10.0)
- python 2.7
Related work:
[1]
@inproceedings{yuan2019simple,
title={A simple convolutional generative network for next item recommendation},
author={Yuan, Fajie and Karatzoglou, Alexandros and Arapakis, Ioannis and Jose, Joemon M and He, Xiangnan},
booktitle={Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining},
pages={582--590},
year={2019}
}
[2]
@inproceedings{yuan2020parameter,
title={Parameter-efficient transfer from sequential behaviors for user modeling and recommendation},
author={Yuan, Fajie and He, Xiangnan and Karatzoglou, Alexandros and Zhang, Liguang},
booktitle={Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval},
pages={1469--1478},
year={2020}
}
[3]
@inproceedings{yuan2020future,
title={Future Data Helps Training: Modeling Future Contexts for Session-based Recommendation},
author={Yuan, Fajie and He, Xiangnan and Jiang, Haochuan and Guo, Guibing and Xiong, Jian and Xu, Zhezhao and Xiong, Yilin},
booktitle={Proceedings of The Web Conference 2020},
pages={303--313},
year={2020}
}
[4]
@article{sun2020generic,
title={A Generic Network Compression Framework for Sequential Recommender Systems},
author={Sun, Yang and Yuan, Fajie and Yang, Ming and Wei, Guoao and Zhao, Zhou and Liu, Duo},
journal={Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining},
year={2020}
}
[5]
@inproceedings{yuan2016lambdafm,
title={Lambdafm: learning optimal ranking with factorization machines using lambda surrogates},
author={Yuan, Fajie and Guo, Guibing and Jose, Joemon M and Chen, Long and Yu, Haitao and Zhang, Weinan},
booktitle={Proceedings of the 25th ACM International on Conference on Information and Knowledge Management},
pages={227--236},
year={2016}
}
[6]
@article{wang2020stackrec,
title={StackRec: Efficient Training of Very Deep Sequential Recommender Models by Layer Stacking},
author={Wang, Jiachun and Yuan, Fajie and Chen, Jian and Wu, Qingyao and Li, Chengmin and Yang, Min and Sun, Yang and Zhang, Guoxiao},
journal={arXiv preprint arXiv:2012.07598},
year={2020}
}
Hiring
If you want to work with Fajie https://fajieyuan.github.io/, Please contact him by email [email protected]. His lab is now recruiting visiting students, interns, research assistants, posdocs, and research scientists. You can also contact him if you want to pursue a Phd degree at Westlake University. Please feel free to talk to him (by weichat: wuxiangwangyuan) if you have ideas or papers for collaboration. He is open to various collaborations. 西湖大学原发杰团队长期招聘：推荐系统和生物信息（尤其蛋白质相关）方向 ，科研助理，博士生，博后，访问学者，研究员系列。