Deep Learning based Recommender System

Note: Some details in the dataset have been obfuscated by the provider in order to maintain the data privacy.

Modern recommendation systems are complicated. The Collaborative Filtering, which used to be the state-of-the-art approach during the Netflix Prize competition, is probably not at the forefront of the research or industry anymore. So this is my motivation to explore a more modern approach, following up from the project that I used to work on a very long time ago when I was in the university.

Recommender systems optimize for different objectives in different contexts. In this dataset, we are interested in predicting the user engagements.

0. System Architecture

1. Metrics

Relative Cross Entropy

Area under the Precision Recall Curve

2. Exploratory Data Analysis

Type

Account creation time

User Followers

Followings

Language

2. Feature Engineering

3. Tree-based Models

4. Deep Learning Models

NCF

Deep Learning Models

References

• Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. In Recsys. 191–198.