DecTag: The Deep Deconfounded Tag Recommender System

The codes are associated with the following paper:

Deep Deconfounded Content-based UGC Tag Recommendation with Causal Intervention,
Yaochen Zhu*, Xubin Ren*, Jing Yi and Zhenzhong Chen

Environment

The codes are written in Python 3.7.12. with the following dependencies.

• numpy == 1.21.2
• pytorch == 1.8.0 (GPU version)
• cudatoolkit == 11.1.1
• scipy == 1.7.3

YT-8M-Causal dataset

The original YouTube-8M dataset can be accessed here.

For preparation, please create a data_split folder.

Examples to run the codes

• Train the deconfounded tag recommender on confounded datasets:

  python train_DecTag_{NFM, LightGCN}.py --dataset YT8M-Causal-{PH, AB} --split [1-5] --gpu [0-7]

  The trained model will be saved in the folder ./check_point/YT8M-Causal-{PH, AB}/{NFM, LightGCN}/split_[1-5].

  (Please create the folder first)

• Evaluate the model and save the testing results:

  python test_DecTag_{NFM, LightGCN}.py --dataset YT8M-Causal-{PH, AB} --split [1-5] --gpu [0-7]

  The results will be saved in the folder ./results/YT8M-Causal-{PH, AB}/{NFM, LightGCN}/split_[1-5].

  (Please create the folder first)

For advanced usage of arguments, run the code with --help argument.

Thanks for your interest in our work.