UniSAR: Modeling User Transition Behaviors between Search and Recommendation

This is the official implementation of the paper "UniSAR: Modeling User Transition Behaviors between Search and Recommendation" based on PyTorch.

Overview

The main implementation of UniSAR can be found in the file models/UniSAR.py.

Experimental Setting

All the hyper-parameter settings of UniSAR on both datasets can be found in files config/UniSAR_KuaiSAR.yaml and config/UniSAR_Amazon.yaml. The settings of two datasets can be found in file utils/const.py.

Quick Start

1. Download data

Download and unzip the processed data Amazon and KuaiSAR. Place data files in the folder data.

2. Satisfy the requirements

The requirements can be found in file requirements.txt.

3. Train and evaluate our model:

Run codes in command line:

python3 main.py --model UniSAR --data KuaiSAR

4. Check training and evaluation process:

After training, check log files, for example, output/KuaiSAR/logs/time.log.

Environments

We conducted the experiments based on the following environments:

CUDA Version: 11.4
OS: CentOS Linux release 7.4.1708 (Core)
GPU: The NVIDIA® 3090 GPU
CPU: Intel(R) Xeon(R) Gold 6230R CPU @ 2.10GHz

TengShi-RUC / UniSAR