Skip to content

Latest commit

 

History

History
39 lines (30 loc) · 2.26 KB

README.md

File metadata and controls

39 lines (30 loc) · 2.26 KB

Data Science Project - Recommender System (Netflix)

Here is my take on the dataset contructed from Netflix which begun in October 2006.
Objective: <0.9525 RMSE score (Root Mean Square Error)

Codes and Resources Used

Python Version: 3.8
Packages: numpy, pandas, matplotlib, wordcloud, sklearn, surprise, tensorflow
Dataset from Kaggle: https://www.kaggle.com/netflix-inc/netflix-prize-data
Prize Details: http://www.netflixprize.com

Data Overview

Dataset was contructed to support the participants in the Netflix Prize.

  • Netflix Customers: 480,000 (randomly chosen and annoymous)
  • Movie Titles: 17,000
  • Data collected from: Oct 1998 to Dec 2005
  • Ratings: 1 to 5 stars
  • Date of Rating
  • Movie ID
  • Movie's Year of Release

Due to the nature of the data provided, collaborative filtering was used in the process. Thereafter, top 3 models were used to recommend the targeted user the next 10 movies to watch.
Netflix Process Overview

Model Building

Due to some restraints, I have reduced the data size from 100million to 5 million. The 5 million data comes from Jun to Dec 2005 since 50% of the data were rated in 2005.
Rating Distribution 1999 to 2005 Rating Distribution 2005

Root Mean Square Error (RMSE) was one of the metrics used to evaluate the project.

Netflix RMSE Overview (tabular format)

Takeaways

Over the span of 10 days, whilst working as a full-timer, I am glad to attain great RMSE scores. I believe with better device and more time given prior to the presentation, I could have better results. As most of my time spent was running the different models with different data size. At the same time, trying to figure out how do I not lose accuracy with less data (from the original dataset provided).