LoID

The source code is for the paper: "Enhancing Content-based Recommendation via Large Language Model" accepted in CIKM 2024 by Wentao Xu, Qianqian Xie, Shuo Yang, Jiangxia Cao, Shuchao Pang.

Requirements

Python=3.11.4

Pytorch=2.0.1

Transformers=4.31.0

CUDA=11.7

Peft=0.5.0

Scikit-learn=1.3.0

Scipy=1.11.1

Preparation

The datasets used in the paper can all be downloaded from this official website.

Usage

To run this project, please make sure that you have the following packages being downloaded.

For LoID:

sbatch LoID.sh

LoID.sh:

#!/bin/bash
#
#SBATCH --job-name=job
#SBATCH --output=output_test%j.txt
#SBATCH --error=errors_%j.txt
#SBATCH --mem=16G
#SBATCH --time=240:00:00
#SBATCH --nodelist=aias-compute-1
#SBATCH --gres=gpu:1

python3 LoID.py

For Pretrain:

sbatch pretrain.sh

Pretrain.sh:

#!/bin/bash
#
#SBATCH --job-name=job
#SBATCH --output=output_test%j.txt
#SBATCH --error=errors_%j.txt
#SBATCH --mem=16G
#SBATCH --time=240:00:00
#SBATCH --nodelist=aias-compute-1
#SBATCH --gres=gpu:1

python3 Pretrain.py

For BERT:

sbatch BERT.sh

BERT.sh:

#!/bin/bash
#
#SBATCH --job-name=job
#SBATCH --output=output_test%j.txt
#SBATCH --error=errors_%j.txt
#SBATCH --mem=16G
#SBATCH --time=240:00:00
#SBATCH --nodelist=aias-compute-1
#SBATCH --gres=gpu:1

python3 BERT.py

For GPT2:

sbatch GPT2.sh

GPT2.sh:

#!/bin/bash
#
#SBATCH --job-name=job
#SBATCH --output=output_test%j.txt
#SBATCH --error=errors_%j.txt
#SBATCH --mem=16G
#SBATCH --time=240:00:00
#SBATCH --nodelist=aias-compute-1
#SBATCH --gres=gpu:1

python3 GPT2.py

For Merging Method DARE：

To run the Multi-Lora module, follow these steps:

1. Navigate to the multi_loras-main\multi_loras directory.
2. Run the merging_methods.py.
3. After obtaining the DARE model, replace the bert-base-cased model in the Loid(DARE) with the DARE model.

python3 merging_methods.py

Then run Loid (DARE):

sbatch Loid(DARE).sh

For Loid(DARE).sh:

#!/bin/bash
#
#SBATCH --job-name=job
#SBATCH --output=output_test%j.txt
#SBATCH --error=errors_%j.txt
#SBATCH --mem=16G
#SBATCH --time=240:00:00
#SBATCH --nodelist=aias-compute-1
#SBATCH --gres=gpu:1

python3 Loid(DARE).py