FILM RECOMMENDATION SYSTEM WITH MATRIX FACTORIZATION METHOD BASED ON COLLABORATIVE FILTERING
The rapid development of the film industry has made film recommendation systems increasingly popular. A film recommendation system is a system that provides a list of film recommendations to users. There are several types of recommendation systems, including knowledge-based recommenders, content-...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/79248 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | The rapid development of the film industry has made film recommendation systems
increasingly popular. A film recommendation system is a system that provides a
list of film recommendations to users. There are several types of recommendation
systems, including knowledge-based recommenders, content-based systems, and
collaborative filtering. In this Final Project, the collaborative filtering type is used
in the matrix factorization method. Matrix factorization is a method that factors the
user-item matrix into two smaller matrices, namely the user matrix and the item
matrix.
This Final Project aims to create and carry out an analysis of the impact of hyperparameters
on the film recommendation system model using the matrix factorization
method. The dataset used in this final assignment is MovieLens 100k
which contains a total of one hundred thousand ratings given by 943 users in 1682
movies. Model training is carried out by varying hyperparameters, namely: number
of features (k), learning rate (?), and regularization factor (?). The randomness
initiation matrix is divided into three initiation matrices.
The hyperparameter value of the number of features (k) determines the level of
complexity of the model, increasing the value of k will improve the model’s performance
in providing predictions. The number of iterations required by a model to
achieve a certain performance can be influenced by the learning rate (?). Choosing
a value of ? that is too large causes the model to not converge. The regularization
factor (?) determines how much penalty is applied to the model so that it can help
the model avoid over-fit. Differences in the initiation of the P and Q matrices in the
training process can affect the resulting model, namely in providing a different list
of recommendations. |
|---|