Artificial Intelligence (AI) and Machine Learning (ML) applications have significantly grown over the years. Most business sectors have acquired, and developed systems driven by ML algorithms. The advancement of these systems allowed huge convenience, not only to the businesses but also to the users. The success of existing ML applications opened several possibilities of future enhancements. More research and experiments are being conducted to elevate the competence of the existing ML methods. In this project, two different approaches in building a Recommender System were evaluated to find out which method will exhibit better performance. Traditional Matrix Factorization and Hybrid Recommender (driven by the newly Neural Collaborative Filtering) were the primary methods used and analyzed during development phase.
Three experiments were conducted to evaluate the performance of each model.
- From the generated graphs, it was demonstrated that RMSE decreases as the number of latent features increases from 20 to 50.
- Top 10 recommendation list was generated from each model. Based on the comparison result, all metrics, except recall, were better in the list generated by Hybrid Recommender.
- Top 20 recommendation list was generated from each model. Based on the comparison result, metrics of the Hybrid Recommender were better than the ones of Matrix Factorization.
In conclusion, Hybrid Recommender exhibited better performance than the other method. Thus, it can be a potential method to improve the traditional recommender systems that drive industry today.
- https://github.com/madserrano (Twitter Analysis)
- https://github.com/iamAjayDahiya (Matrix Factorization)
- https://github.com/jonatasaguiar (Hybrid Recommender)