PRM

Code for the paper "Personalized Re-ranking for Recommendation"

Dataset

the toy data for training/validation/test

• rec_train_set.sample.data-->8000 lines
• rec_validation_set.sample.data-->1000 lines
• rec_test_set.sample.data-->1000 lines

Training

Without Personalized Vector

python exec.py \
        --train true \
        --train_set dataset/rec_train_set.sample.txt \
        --validation_set dataset/rec_validation_set.sample.txt \
        --batch_size 512 \
        --train_epochs 100 \
        --train_steps_per_epoch 1000 \
        --validation_steps 2000 \
        --early_stop_patience 10 \
        --lr_per_step 4000 \
        --d_feature 19 \
        --pos_embedding_mode 0 \
        --saved_model_name PRM_no_pv.h5

With Personalized Vector

python exec.py \
        --train true \
        --train_set dataset/rec_train_set.sample.txt \
        --validation_set dataset/rec_validation_set.sample.txt \
        --batch_size 512 \
        --train_epochs 100 \
        --train_steps_per_epoch 1000 \
        --validation_steps 2000 \
        --early_stop_patience 10 \
        --lr_per_step 4000 \
        --d_feature 19 \
        --saved_model_name PRM_pv.h5

Testing

python exec.py \
        --test_set dataset/rec_test_set.sample.txt \
        --validation_set dataset/rec_validation_set.sample.txt \
        --batch_size 512 \
        --saved_model_name PRM_pv.h5

Metric Evaluation

python metric.py dataset/rec_test_set.sample.txt.predict.out