RecommenderSystem

Recommendation System Code Repository

This code repository contains implementations of various recommendation algorithms and evaluation tools, supporting multiple datasets and evaluation metrics.

Directory Structure

├── base_recommender.py # Base class for all recommender algorithms

├── models/

│ ├── user_cf.py # User-based Collaborative Filtering

│ ├── itemcf.py # Item-based Collaborative Filtering

│ ├── baseline.py # Baseline models (Random and Popularity recommenders)

│ ├── content_based.py # Content-based Recommendation

│ ├── friend_based.py # Friend-based Collaborative Filtering

│ ├── SVD.py # Matrix Factorization with SVD

│ ├── NGCF.py # Neural Graph Collaborative Filtering

│ ├── LGCN.py # LightGCN Graph Neural Network

│ └── init.py

├── data_processor.py # Data processing utilities

├── evaluation.py # Evaluation metrics and tools

└── experiment.py # Main experiment script

Code File Descriptions

base_recommender.py

Defines the base class for all recommender algorithms, providing a unified interface with fit and recommend methods.

models/user_cf.py

Implements user-based collaborative filtering, recommending items by calculating user similarities.

models/itemcf.py

Implements item-based collaborative filtering, recommending items by calculating item similarities.

models/baseline.py

Contains two baseline recommendation algorithms:

• RandomRecommender: Recommends items randomly.
• PopularityRecommender: Recommends items based on their popularity.

models/content_based.py

Content-based recommendation algorithm that uses item feature information (e.g., movie genres) for recommendations.

models/friend_based.py

Collaborative filtering algorithm that incorporates user friendship relationships for recommendations.

models/SVD.py

Matrix factorization-based recommendation algorithm using Singular Value Decomposition (SVD).

models/NGCF.py

Neural Graph Collaborative Filtering algorithm that captures high-order user-item relationships using graph neural networks.

models/LGCN.py

LightGCN algorithm, an efficient graph neural network-based recommendation method.

data_processor.py

Utilities for data loading, preprocessing, and splitting, supporting multiple datasets.

evaluation.py

Tools for evaluating recommender systems, providing multiple metrics (e.g., HR@K, NDCG@K, MRR@K, Recall@K).

experiment.py

Main script for running experiments with different recommendation algorithms.

Supported Datasets

• MovieLens-1M (ml-1m)
• Last.fm (lastfm)
• Yelp (yelp)

Experiment Command-Line Scripts

MovieLens-1M Dataset

UserCF Model

python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model usercf --k 20 --split_method default
python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model usercf --k 30 --split_method default
python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model usercf --k 40 --split_method default

#### Data Splitting Methods
python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model usercf --k 30 --split_method time
python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model usercf --k 30 --split_method random

#### Other Models
python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model svd --n_factors 500 --n_epochs 300 --split_method default
python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model lightgcn --embed_dim 256 --n_layers 4 --epochs 5 --split_method default
python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model ngcf --embed_dim 64 --n_layers 3 --epochs 5 --split_method default
python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model random --split_method default
python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model popular --split_method default
python experiment.py --dataset ml-1m --data_path /path/to/ml-1m/ --model content --split_method default

#### Last.fm Dataset
python experiment.py --dataset lastfm --data_path /path/to/lastfm/ --model usercf --k 20 --split_method default
python experiment.py --dataset lastfm --data_path /path/to/lastfm/ --model itemcf --k 20 --split_method default
python experiment.py --dataset lastfm --data_path /path/to/lastfm/ --model svd --n_factors 200 --n_epochs 500 --split_method default
python experiment.py --dataset lastfm --data_path /path/to/lastfm/ --model lightgcn --embed_dim 128 --n_layers 2 --epochs 20 --split_method default
python experiment.py --dataset lastfm --data_path /path/to/lastfm/ --model ngcf --embed_dim 64 --n_layers 3 --epochs 20 --split_method default
python experiment.py --dataset lastfm --data_path /path/to/lastfm/ --model random --split_method default
python experiment.py --dataset lastfm --data_path /path/to/lastfm/ --model popular --split_method default

#### Yelp Dataset
python experiment.py --dataset yelp --data_path /path/to/yelp/ --model usercf --k 20 --split_method default
python experiment.py --dataset yelp --data_path /path/to/yelp/ --model svd --n_factors 200 --n_epochs 500 --split_method default
python experiment.py --dataset yelp --data_path /path/to/yelp/ --model lightgcn --embed_dim 128 --n_layers 2 --epochs 20 --split_method default
python experiment.py --dataset yelp --data_path /path/to/yelp/ --model ngcf --embed_dim 64 --n_layers 3 --epochs 20 --split_method default
python experiment.py --dataset yelp --data_path /path/to/yelp/ --model random --split_method default
python experiment.py --dataset yelp --data_path /path/to/yelp/ --model popular --split_method default

### Notes
The MovieLens dataset supports the time splitting method, while Last.fm and Yelp datasets do not.
Ensure all dependencies are installed and adjust data paths as needed.