Source code of WWW 2022 Paper "Modeling User Behavior with Graph Convolution for Personalized Product Search".
- environment
conda env create -f SBG/air.yml
-
prepare data
- download amazon data from https://jmcauley.ucsd.edu/data/amazon/
- specify the input and output dir in
preprocessing/prepare_amazon.py
- run
preprocessing/prepare_amazon.py
for corresponding datasets - config dataset in
persearch.config.cfg_data
-
reproduce the results
cd SBG
python main.py -d amazon_software@ -e 200 -r 5
-
configuration
- model configuration in
persearch.config.cfg_model
- training data generator configuration in
persearch.config.cfg_gen
- other command line interactions:
persearch.args
- model configuration in
-
customize model
- training data generator for the model
- build Generator pipeline in
persearch.gen
- register gen in
persearch.config.cfg_gen
, add its name in yourpersearch.config.cfg_model
with the key'generator'
- build Generator pipeline in
- implement
forward, do_train, and f_loss
- doc refer to
persearch.model.Base
- example as
persearch.model.zam.ZAM
- example as
- register model in
persearch.config.cfg_model
- load it in
model/__init__.py
- write test in
exps.py
- training data generator for the model
-
log
- summary stored in
logs/<dataset>/<dataset_ver>/<arg.caption>/<timestamp>
- summary stored in