This repository contains solution for Sony R.I.S.E. Challenge. The solution required creating a Content Recommendation System for suggesting top contents to watch for a particular user.

You can find the analysis, ideation and description of the final recommender system created below.

• For creating a recommendation system, we focused on algorithms such as Collaborative Filtering and Content Based Filtering. (Collaborative Filtering works by filtering out items that a user might like based on reactions by similar users.)
• Since we didn't have individual content ratings given by each user to the content they have watched, we did not focus much on User based Collaborative Filtering
• We focused on recommending content to the users based on their watch history. For this we needed to find a way to decide which content are similar and combine their choices to create a list of recommendations.

• To compare content, we reduced the content feature data to embeddings, by training a supervised Autoencoder and a Predictor DNN.
• The embeddings map the feature data into a space with lower dimensions and generally capture some latent structure of feature dataset.
• We extracted the generated embeddings and used them to calculate similarity between any two content.
• We used cosine similarity and manhattan distance to compare embeddings. We found cosine similarity to work a little better in our case.
• Once we could find similarity between contents, we just used a user's previous watch history to recommend Top 10 content similar to the same.

You can refer to the jupyter notebook for the solution.