KNN Recommendations — Part 6

Table of Contents :

1. Introduction and Recommendation Framework
2. Evaluating Recommendation Systems
3. Content Based Recommendations
4. Neighborhood Based Collaborative Filtering
5. User and Item Based Collaborative Filtering
6. KNN Recommendations
7. Matrix Factorisation
8. Deep Learning — Introduction
9. Restricted Boltzmann Machines
10. AutoRecs
11. Amazon DSSTNE and Sage Maker
12. Real-World Challenges and Solutions

The architecture below showcases collaborative filtering applied to the recommendation system for “Rating Predictions”.

This is an efficient approach!

But since we are predicting rating values, we can measure offline accuracy of the system using Test/Train or Cross-Validation; which is useful in the research world!

User-Based KNN Recommendations

• Still uses a user-user similarity matrix at its core
• We are reaching deeper into data and trying to make rating predictions
• Complex approach — Which is a bad thing!!

Item-Based KNN Recommendations

• We are better off with a simpler approach rather than a complicated one
• This is more like shoveling into the framework

Above all, offline metrics cannot be the pure baseline for choosing one framework.

TRANSLATION BASED RECOMMENDATIONS(2017)

We embed items into a ‘transition space’ where users are modeled as translation vectors operating on item sequences. For example, if the user watched a Tom Cruise movie and then watches his movie again and then again watches his movies, it is an effective approach to recommend Tom Cruise’s movie to the user.

• Outperformed all other methods (Hit Rate)
• Under research(Promising)
• Code and data available online

NEXT>>>