A-RecSys implements a set of implicit recommendation algorithms and is designed to be used in large scale recommendation settings. It extends traditional matrix factorization algorithms and focuses on attribute embedding and applying sequence models.
Works implemented by this toolkit include:
- A Batch Learning Framework for Scalable Personalized Ranking. AAAI 18. arXiv
- Sequential heterogeneous attribute embedding for item recommendation. ICDM 17 SERecsys Workshop.
- Temporal Learning and Sequence Modeling for a Job Recommender System. RecSys Challenge 16 pdf
Currently the models and features supported by A-RecSys include,
- Hybrid matrix factorization model (with deep layer extensions)
- Linear sequence models based on CBOW and skip-gram
- LSTM-based seq2seq model
- Recommendation with implicit feedback
- Heterogeneous attribute embedding (see attributes/README.md for details)
- Objective functions include cross-entropy, Weighted Margin Rank Batch loss.
CSV-formated (sep=\t) input files include
u.csv : user file. user id and attribute values.
i.csv: item file. item id and attribute values.
obs_tr.csv: implicit feedback for training. First two columns are user-id, item-id. Third column (optional) is for timestamp.
obs_va.csv: implicit feedback for development.
obs_te.csv: implicit feedback for testing.
A example (adapted from MovieLens 1m) is given at ./examples/dataset/
Example scripts are provided at ./examples/ to start running the code
To train hybrid matrix factorization model on provided MovieLens 1m dataset,
cd examples/
bash run_hmf.sh 32 1 False 100 False
To train lstm model,
cd examples/
bash run_lstm.sh 64 1 False
(Code has been tested on TF 0.8 and above.)
You can switch to "recommend" mode from "training" by setting flag recommend to 'true'. In the above HMF example, it would be
cd examples/
bash run_hmf.sh 32 1 False 100 True
By default, the code generates a ground truth interaction file res_T_test.csv from obs_te.csv, and perform recommendation on all users that appear in res_T_test.csv; however, you can generate your own res_T_test.csv to narrow down user set on which recommendation is performed.
The code now supports Tensorflow v1.0. During our development, code is tested with versions 0.8, 0.9, 0.11, 0.12.
Please cite the following if you find this helpful.
@inproceedings{liu2017wmrb, title={WMRB: learning to rank in a scalable batch training approach}, author={Liu, Kuan and Natarajan, Prem}, booktitle={Proceedings of the Recommender Systems Poster}, year={2017}, organization={ACM} }
@inproceedings{liu2016temporal, title={Temporal learning and sequence modeling for a job recommender system}, author={Liu, Kuan and Shi, Xing and Kumar, Anoop and Zhu, Linhong and Natarajan, Prem}, booktitle={Proceedings of the Recommender Systems Challenge}, pages={7}, year={2016}, organization={ACM} }
Your comments and suggestions are more than welcome! We really appreciate that!
Kuan Liu [email protected] Xing Shi [email protected]