Authors : Zihe Wang (zw2624), Di Ye (dy2404), Ziyao Zhang (zz2583), Yinhe Lu (yl4372)
Date: Dec 20, 2019
Please install the required package using pip install -r requirements.txt. (Package of Factorization Machine can be installed using the link in reference below.)
Directory Code contains developing codes in jupyter notebook.
Directory Notebooks contains generated PDFs of jupyter notebooks we used to produce our report.
Directory Data contains the small datase,(large datsest is too big to upload).
Please see final report.pdf for summarization of everything above and further evaluation.
Our business objective is to predict rating of user to business and rank the predicted business and recommand them to the users, and we choose those users who have rated at least few businesses on the platform. We want to make sure that we can accurately predict the ratings and recommend the businesses to user in the order they would prefer.
Summary of the objectives:
- How close our prediction is to the actual rating
- Whether the predicted ranking of businesses is the same as the actual ranking
- How do our models perform on different segmentation of user and business
Full dataset can be found here: Yelp dataset challenge https://www.yelp.com/dataset/challenge.
We construct a large dataset and a smaller dataset to work on.
We also segment the large test set from user and business dimension by three different popularities.
Please see final_report.pdf
Recommender Systems: The Textbook, By Charu C. Aggarwal
Lecture Notes from IEORE 4571, by Dr. Brett Vintch, Columbia University
Neural Collaborative Filtering: https://arxiv.org/abs/1708.05031
Factorization Machines in Python: https://github.com/coreylynch/pyFM
Wide & Deep Learning for Recommender Systems, Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen: https://arxiv.org/pdf/1606.07792.pdf