-
Notifications
You must be signed in to change notification settings - Fork 0
/
Documentation.txt
19 lines (10 loc) · 1.22 KB
/
Documentation.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Please read the confluence page first to get the idea of the project.
First two models (Nearest neighbour and Latent factor) were implemented for book recommendations in python.The folder Nearest_Neighbor&Latent_Factor folder includes code Book_recommendation.py and the dataset can be found at:
http://www2.informatik.uni-freiburg.de/~cziegler/BX/
The deep learning models(RBM and Autoencoders) were implemented on movielens data(ml-100k) for movie recommendation. Dataset can be found at:
https://grouplens.org/datasets/movielens/
The implemented RBM model is binary model whether the user will like the movie or not.
The implemented Autoencoder model has discrete output that is prediction for ratings in 1-5 range. (The loss can be minimised further by increasing the layers and tweaking the model)
The folder RBMs contains data and code for RBM model and the autoencoders folder contains the data and code for Autoencoder model.
Conclusion:
The Netflix challenge winning solution used Latent factor implementation which proves to be best. In the code I have written you’ll get the reconstructed matrix and the task is just to sort and predict. The auto encoder model has a loss of 0.7 so +- 0.7 in rating prediction is expected.