DefinePK

Research Objective

To develop and evaluate a Weighted Hybrid Model that combines Content-Based Filtering, Collaborative Filtering, and Popularity-Based recommender systems to improve recommendation accuracy and address the cold-start problem.

Methodology

The research proposes a Weighted Hybrid Model that linearly combines the scores of three standalone recommender models: Content-Based Filtering, Collaborative Filtering, and Popularity-Based. The model was developed in Python using a cross-validation technique (Holdout) with 80% of the data for training and 20% for testing. A Kaggle dataset containing 12 months of news articles and user interactions was used. Data pre-processing involved assigning weights to different user interaction types (View, Like, Bookmark, Follow, Comment) and removing users and interactions with low weights. The performance was evaluated using Top-N accuracy metrics, specifically Recall@N.

Methodology Flowchart

                        graph TD;
    A[Data Collection: News Dataset] --> B[Data Pre-processing: Interaction Weighting & Filtering];
    B --> C[Model Development: Hybrid Model];
    C --> D[Combine Content-Based, Collaborative, Popularity Scores];
    D --> E[Model Training];
    E --> F[Model Evaluation: Recall@N];
    F --> G[Results Analysis];
    G --> H[Conclusion & Implications];                    

Discussion

Standalone recommender systems like Content-Based Filtering and Collaborative Filtering have limitations such as cold-start and plasticity. The proposed hybrid approach effectively mitigates these issues by combining the strengths of multiple models. The integration of a popularity-based model further enhances accuracy and addresses the cold-start problem by leveraging trending items. The research suggests that this hybrid approach can be extended beyond news recommendations to other domains like e-commerce.

Key Findings

The proposed hybrid model, when combined with the popularity-based model, achieved a Recall@10 of 90%. The best-performing standalone model achieved an efficiency (Recall@10) of 53%. The hybrid model demonstrated higher accuracy compared to individual Content-Based, Collaborative, and Popularity-Based models.

Conclusion

A Weighted Hybrid Model combining Content-Based Filtering, Collaborative Filtering, and Popularity-Based recommender systems was successfully developed. This hybrid model significantly improved recommendation accuracy, achieving 90% Recall@10, outperforming individual models. The research highlights the effectiveness of hybridization in addressing the limitations of standalone recommender systems and suggests its applicability to various recommendation domains.

Fact Check

- The hybrid model combined with the popularity-based model achieved a Recall@10 of 90%.
- The best-performing standalone model achieved an efficiency (Recall@10) of 53%.
- The dataset used contained data from approximately 73,000 logged-in users and over 3,000 user interactions before pre-processing.

Mind Map