SEAR: LLM-Powered Sequential Recommendation via Fusion of Collaborative, Semantic, and Rating Information

Implementation of "SEAR: LLM-Powered Sequential Recommendation via Fusion of Collaborative, Semantic, and Rating Information".

Datasets

The statistics of datasets used in the experiments.

	# Users	# Items	# Iterations	Sparsity	Avg. Iter$_{user}$	Avg. Iter$_{item}$
Beauty	22,363	12,101	198,502	99.93%	8.88	16.40
Games	54,955	17,256	494,934	99.95%	9.01	28.68
Supplies	236,506	42,351	2,093,662	99.98%	8.85	49.44
Scientific	10,646	4,991	74,357	99.86%	6.98	14.90
ML-1M	6,040	3,416	999,611	95.16%	165.50	292.63

Using Our Code to Reproduce the Results

1. Set Up the Environment.

• Software Requirements:

  • Python: 3.10.16
  • CUDA: 11.8

• Choose one of the following methods to install dependencies:

--- Option 1: Install from requirements.txt

 conda install --yes --file requirements.txt # You may need to downgrade the torch using pip to match the CUDA version

--- Option 2: Install packages individually

conda create -n SEAR python=3.10
pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu118
pip install recbole==1.2.0
pip install causal-conv1d==1.1.3.post1
pip install mamba-ssm==1.1.4
pip install sentence_transformers==3.2.1
pip install transformers==4.46.3

2. Download open-source LLM all-mpnet-base-v2

   ├── all-mpnet-base-v2
   │ ├── 1_Pooling
   │ │  ├── config.json
   │ ├── config.json
   │ ├── config_sentence_transformers.json
   │ ├── modules.json
   │ ├── README.md
   │ ├── sentence_bert_config.json
   │ ├── special_tokens_map.json
   │ ├── tokenizer.json
   │ ├── tokenizer_config.json
   │ ├── vocab.txt
   │ └── model.safetensors

3. Download dataset

• Download dataset from here.

   ├── dataset
   │ ├── amazon-beauty
   │ │ ├── amazon-beauty.inter
   │ │ ├── amazon-beauty.item
   │ ├── Amazon_Video_Games
   │ │ ├── ...
   │ ├── Amazon_Pet_Supplies
   │ │ ├── ...
   │ ├── Amazon_Industrial_and_Scientific
   │ │ ├── ...
   │ ├── ml-1m
   │ │ ├── ...

• Set the following variations in config/data/~.yaml

data_path: # i.e.,  '/mnt/public/gw/RecLLM/dataset/'

4. Train and test our proposed deep model.

• Run python custom_trainer.py from the root directory.