This is a Python 3 based collaborative filtering recommendation system based on Singular Value Decomposition (SVD). The module allows users to load from a CSV or from a numpy array/matrix. This is converted to a sparse matrix, and SVD is computed to convert the user/item matrix into a k-dimensional space where recommendations are computed based on vector overlaps.
With this system you can pull out recommendations for users, find similar items, find users with similar history, and make recommendations based on user similarity.
An example of using this system can be found here: Movie Lens Recommender. In short, given a user item matrix this module can reliably find similar items based on user input. It can also reliably find items that a user will be interested in based on their history with other items in the set.
Below, the ID for Toy Story 2 is input and the most similar items are Toy Story, A Bug's Life, Who Framed Roger Rabbit?, and Finding Nemo. All of these are animated children's movies, with the majority being Pixar movies.
MOVIE_ID = 3114 # Toy Story 2
for item in svd.get_similar_items(MOVIE_ID,show_similarity=True):
print(item)
print(svd.get_item_name(item[0]),'\n')(3114, 0.12452773823527524)
Toy Story 2 (1999)
(1, 0.096984857294616089)
Toy Story (1995)
(2355, 0.043104443630875205)
Bug's Life, A (1998)
(2987, 0.041949127538017023)
Who Framed Roger Rabbit? (1988)
(6377, 0.040522854363774369)
Finding Nemo (2003)
To install, just do pip install svdRec from the command line. This will also
install numpy and scipy since they are required.
To use, you have to do (it's currenly ugly, but I'm working on patching it for v0.2):
from svdRec import svdRec
svd = svdRec.svdRec()From there, all the features are just as shown in the example notebook.