Deep Learning — Part 8

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

In this section we won’t be talking about Deep Learning but rather we will be answering questions such as ‘can we use DL for our recommendation sysyems?’ and how good the approach will be.

Image source: (machinelearningmastery.)

IS DEEPLEARNING A OVERKILL?

Just because new tech is hot, it doesn’t mean it is right solution for every problem!

DL is shows good performance in

• Actions similar to our brain
• Image recognition
• Series/Pattern recognition

Neural nets are fundamentally matrix operations (which is proved by Tensorflow). And there are already well established
matrix factorisation techniques for recommendation systems that fundamentally do something similar.
For example, in SVD, We find matrices using weights calculated by SGD; Which is similar to DL

Hence, DL CAN BE APPLIED TO RECOMMENDATION SYSTEMS

• It is just some kind of pattern recognition where people are involved.
• Amazon open-sourced neural nets based DSSTNE (where huge neural nets deal with sparse data)
• Even though by small margins, in some researches DL outperformed SVD

NEXT >>>