Matrix Factorization : How is it used in recommender systems? | Breakthrough Junior Challenge 2020

so you've just finished watching a series on Netflix and once again it's time for you to make that overwhelming decision of what to watch next as he was scrolling through the screens of Netflix I'm sure you've noticed this and you may have wondered what these numbers even mean well I'm sure I can explain [Music] yes matrix factorization as the name suggests the idea behind it is to decompose a specific matrix into discrete matrices primarily to ease analysis have a focused representation and to save storage space picturise the data set of thousand movies and 400 different users without any form of decomposition it would take over 400,000 different entries when introduce simple decomposition factors called Latin factors and one can reduce those significantly to just 7,000 I know still sounds like a lot well introduced complex decomposition models and factors like this and one can subsequently funnel it down further in the years 2006 through 2009 a competition Netflix prize was held which intended to award a million dollars to the participants who could improve Netflix is original recommender system the most commonly used matrix factorization technique by all of the participants in the competition was singular value decomposition consider this formula it represents the basic mathematical expression for singular value decomposition hey operand matrix a is shown to be equal to the sum of the user into rank matrix rank into item matrix and the key matrix Sigma that provides user bytes more graphically one can show pair in matrix a to be the sum of multiple individual products of these matrices however big data bases like that of Netflix's no need to update their mattresses or addition of a new user machines is in Sigma are almost negligible another more common matrix factorization technique used these days is a deeper decomposition model which uses both Xmas interactions like readings and implicit interactions like individual user bookmarks watch or skip videos even Amazon ecommerce prime video and your daily mix on Spotify make extensive use of such recommender systems one can even relate this to basic Association rules followed in the tourism industry for instance a select group of tourists tend to travel to nearby destinations all in a single trip as in when tourists from India or travel to Singapore they're often seen adding Malaysia to the itinerary as well such innate abilities of math are what make it universal it is what has made it possible for us to keep count of the length of this video and it will probably fuel our next expedition to the proximal galaxy I am Mike Rowe and thank you for watching

Channel: Mayank Roy

Views: 952

Rating: 4.9607844 out of 5

Keywords: breakthrough, matrix factorization, factorisation, recommender system, netflix prize, singular value decomposition, deeper decomposition, machine learning, artificial intelligence, netflix, youtube, spotify daily mix, matrices, gradient descent, latent factor, breakthrough junior challenge 2020, association rules

Id: g_buOuTr1fY

Length: 3min 0sec (180 seconds)

Published: Tue Jun 23 2020