Solution of SIGIR Ecom Data Challenge 2021

Overview

Coveo hosts the 2021 SIGIR eCom Data Challenge

Requirements

• torch==1.8.1

• tqdm==4.60.0

• numpy==1.20.2

• boto3==1.15.8

• python-dotenv==0.13.0

Getting Started

mkdir saved log results
mkdir dataset & cd dataset
mkdir new prepared raw

The path of raw dataset is ./dataset/raw

Pre-Process

Run the scripts in ./scripts

Models

txt embedding

 python train.py --model GRU4Rec --device 0 --lr 1e-4 --seq_mode sku --commit txt
 python train.py --model GRU4Rec --device 0 --lr 1e-4 --seq_mode sku --commit txt --evaluate

deepwalk embedding(url-sess-item)

 python train.py --model GRU4Rec --device 0 --lr 1e-4 --seq_mode sku --commit dw
 python train.py --model GRU4Rec --device 0 --lr 1e-4 --seq_mode sku --commit dw --evaluate

deepwalk embedding(item-item)

 python train.py --model GRU4Rec --device 0 --lr 1e-4 --seq_mode sku --commit dw_i-i
 python train.py --model GRU4Rec --device 0 --lr 1e-4 --seq_mode sku --commit dw_i-i --evaluate

rand embedding

 python train.py --model GRU4Rec --device 0 --lr 1e-4 --seq_mode sku --commit rand
 python train.py --model GRU4Rec --device 0 --lr 1e-4 --seq_mode sku --commit rand --evaluate

Post-Process

Run ensemble.ipynb