Overview / Usage

In an age where trying to make a decision due to so many options available, having a recommender system has become essential. Netflix, YouTube, Amazon, and so many other businesses use this to the optimum.

The point of creating this recommender is to allow developers to take this and build their own recommender systems using different datasets, and use this as a base recommendation system. It can easily be plugged into applications to do recommendations. On its own though, this is a recommendation system for Movies. So we can pass in user ids, the model, and the data (movies in this case) to create recommendations.

I used the movie datasets provided by LightFM to predict and recommend the top 3 movies in the list based on a user's past ratings and selections, as well as what other similar users selected.

Methodology / Approach

Used LightFM to get the model, and create it. Then trained the model using the fit method from the lightFM library. LightFM assumes that ratings 4 or below are negative and the rest are positive.

Gradient descent algorithm to iteratively find the weights and improve predictions over time. This takes into consideration the Content that was selected in the past and its Collaborative based on similar users ratings. Which makes this a Hybrid System.

Knowing that there are 2 kinds of recommendation systems: content-based & collaborative based, I chose to go with a hybrid using the lightFM library.

Technologies Used

Intel distribution for Python, lightFM, numpy
Entire structure on Anaconda environment

Repository

https://github.com/akhilboddu/recommedationSystem