FLIP: Fine-grained Alignment between ID-based Models and Pretrained Language Models for CTR Prediction

This is the pytorch implementation of FLIP proposed in the paper FLIP: Fine-grained Alignment between ID-based Models and Pretrained Language Models for CTR Prediction. (RecSys'24)

Requirements

pip install -r requirments.txt

First Step: pretrain FLIP

To pretrain FLIP, please run run_script.py

Second Step: finetune FLIP

To finetune the ID-based model ($\text{FLIP}_{ID}$), please run finetune_ctr.py

To finetune the PLM-based model ($\text{FLIP}_{PLM}$), please run finetune_nlp.py

To finetune both the ID-based model and PLM-based model (FLIP), please run funetine_all.py

How to run baselines

We also provide shell scripts for baselines.

To run the ID-based model baseline:

python ctr_base.py

To run the PLM-based model baseline:

python ctr_bert.py
python ptab.py