Skip to content

Implementing and comparing various techniques for building a  Recommender System. 

Notifications You must be signed in to change notification settings

shrishtrip/Recommender-System

Repository files navigation

Recommender-System

Implementing and comparing various techniques for building a  Recommender System. 

---Recommender System using various approaches.---

Loosely speaking, there are 2 broad ways:

  1. Nearest Neighbour Approach
  2. Find the k (as required) nearest neighbours to a given point. Then the characteristics of that point can be approximated to some function of the characteristics of the neighbours. This is quite intitive, but not always necessarily correct.
  3. Factorization Approach
  4. As the Netflix Prize competition has demonstrated, matrix factorization models are superior to classic nearest-neighbor techniques for producing product recommendations, allowing the incorporation of additional information such as implicit feedback, temporal effects, and confidence levels.

Packages required to run the code:

  • numpy
  • pandas
  • math
  • time
  • scipy
  • csv

After downloading the above packages, download this folder and make it the working directory. Then run the corresponding files for the differnet approaches:

  • IIcollaborative.py-- for Item-Item Collaborative filtering
  • IIcollaborativeWithBaseLine.py --for Item-Item Collaborative filtering with baseline approach
  • UUcollaborative.py.py --for User-User Collaborative filtering
  • UUcollaborativeWithBaseLine.py.py --for User-User Collaborative filtering with baseline approach
  • Latent Factor Mosel.py --for Latent Factor Model

The comparison for all these in mentioned in the design doc. Dataset was taken from https://grouplens.org/datasets/movielens/

About

Implementing and comparing various techniques for building a  Recommender System. 

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages