Deep multi-Representational Item NetworK for CTR prediction

Note: we use Python 2.7 and Tensorflow 1.4.

Prepare Data

./prepare_data.sh

When you see the files below, you can do the next work.

• cat_voc.pkl
• mid_voc.pkl
• uid_voc.pkl
• local_train_sample_sorted_by_time
• local_test_sample_sorted_by_time
• reviews-info
• item-info

Train Model

mkdir dnn_best_model

CUDA_VISIBLE_DEVICES=0 python ./script/train.py train [model name]

The model blelow had been supported:

• SVDPP
• Wide (Wide&Deep NN)
• DIN
• DIEN (https://github.com/mouna99/dien)
• DMIN (https://github.com/mengxiaozhibo/DMIN)
• DUMN (https://github.com/hzzai/DUMN)
• DRINK

Acknowledgements

Our code is implemented based on DIEN, DMIN-CIKM20, TIEN-CIKM20, and DUMN-SIGIR21.