Model-based Collaborative Filtering
Pratheerth Padman
Freelance Data Scientist
Module
Overview
Understanding model-based collaborative
filtering
Demo: Movie recommendations with SVD
Demo: Book recommendations with deep
learning
Understanding Model-based
Collaborative Filtering
Model-based Collaborative Filtering
Model-based collaborative filtering uses machine learning and data
mining capabilities to discover patterns and correlations and come up
with predictions.
Model-based Collaborative Filtering Approaches
Nonparametric KNN
Models that do not rely on any
specific parameter settings
Model-based
collaborative
filtering
K-nearest Neighbors
Say you want to classify Start by calculating the Next, find the nearest Vote on the predicted
the grey point into a distances between the neighbors by ranking class labels based on the
class. grey point and all other points by increasing classes of the k nearest
Here, there are three points. distance. neighbors. Here, the
potential classes – blue, The nearest neighbors labels were predicted
green and orange. (NNs) of the grey point based on the k=3 nearest
are the ones closest in neighbors.
dataspace.
Model-based Collaborative Filtering Approaches
Nonparametric KNN
Singular Value
Decomposition
Model-based Probabilistic
Matrix
collaborative Matrix
factorization Factorization
filtering
Breaking down one matrix into Non-ve Matric
a product of multiple matrices Factorization
Singular Value Decomposition
User-item matrix tends to be very sparse
SVD is a matrix factorization technique that is usually used to reduce
the number of features of a dataset
Datamxn = Umxm Σmxn VTnxn
Latent factor describes a property or concept of a user or item
Singular Value Decomposition
ˆ∑ Vˆ T
rxr rxd
A Û
nxd = nxr
U ∑ V T
nxn nxd dxd
Model-based Collaborative Filtering Approaches
Nonparametric KNN
Singular Value
Decomposition
Model-based Probabilistic
Matrix
collaborative Matrix
factorization Factorization
filtering
Non-ve Matric
Factorization
Neural
Deep learning collaborative
filtering
Embeddings
An embedding is a low-dimensional space that captures the relationship of vectors from a
higher dimensional space.
Likes action movies
+1
Bob Joe
-1 +1 Likes romance movies
-1
User Embeddings
Bob (UserID 1)
() 1
0
() 0.8
User embedding
..
-0.3
0
Joe (UserID 2)
() 0
1
..
0 () 0.9
0.8
One-hot encoded Embedded
representation of users representation of users
(N-dimensional vector) (2-dimensional vector)
Model Architecture
User
Combine
Output
Item
Input Embedding layers Hidden layers
Demo
Movie recommendations with SVD
Demo
Book recommendations with deep learning
Model-based collaborative filtering uses
machine learning and data mining
Summary capabilities to discover patterns and
correlations
3 main types:
- Nonparametric
- Matrix factorization
- Deep learning
Learned to code recommender systems
using SVD and deep learning
Discussion tab for any feedback or questions
Feedback Leave a star rating!
Thank you for watching!