tuansunday05 / ArticleRecommenderSystem

Using hybrid recommender system with apriori algorithm, content-based and collaborative filtering method for predicting users interactions and then recommend them for users.

Hybrid Recommendation System

Welcome to the repository for our hybrid recommendation system, featuring collaborative filtering with Singular Value Decomposition (SVD), content-based filtering using TF-IDF with a vectorizer, and association rule mining through the Apriori algorithm. This system not only combines diverse recommendation approaches but also includes real-time update functions for on-the-fly user additions, making it suitable for server environments demanding instant responses.

Introduction

Our recommendation system aims to deliver personalized suggestions by seamlessly integrating collaborative filtering, content-based analysis, and association rule mining. The unique feature of this system lies in its ability to adapt to dynamic user profiles in real-time, making it well-suited for applications where user interactions change rapidly.

Components

Popularity Model

• Recommends popular items to users based on overall item popularity.
• Provides a simple yet effective baseline for comparison with personalized recommendation approaches.

Association Rule Mining (Apriori Algorithm)

• The Apriori algorithm is used to discover association rules among items.
• This component aims to capture implicit relationships between items and improve the diversity of recommendations.

Collaborative Filtering (SVD)

• Singular Value Decomposition (SVD) is used to factorize the user-item interaction matrix, capturing latent factors that represent user and item preferences.
• Implementation using the Surprise library for collaborative filtering.

Content-Based Filtering (TF-IDF with Vectorizer)

• TF-IDF (Term Frequency-Inverse Document Frequency) is employed to represent the content features of items.
• A vectorizer is used to convert textual information into numerical features.
• This approach allows the system to recommend items based on their content similarity.

Hybrid Model

• The recommendations from the collaborative filtering, content-based filtering, and association rule mining components are combined to form the hybrid model.
• The final recommendations are generated by considering the strengths of each individual model.

Real-Time User Updates

• Using update_user_profile method of HybridRecommender class with the ID of user needed to update information.

Results

• Evaluation metrics and comparisons of the hybrid model against individual models are provided.
• The hybrid model is demonstrated to outperform single models in terms of recommendation accuracy and coverage.
• Real-time user updates contribute to the adaptability and responsiveness of the recommendation engine.

Usage

1. Clone the repository:

    git clone https://github.com/tuansunday05/ArticleRecommenderSystem.git
   

2. Go to root folder:

   cd ArticleRecommenderSystem/
   

3. Install dependencies

    pip install -r requirements.txt
   

4. Run the recommender engine:

    python3 scripts/models/hybrid_developing.py
   

License

This project is licensed under the MIT License.