A Recommender System Supporting Multiple Algorithms

Team members : Sewon Min, Chaofan Han

Project Abstract

In the current era of information explosion, recommender systems gain more and more popularity as they can make appropriate and customized recommendations to users based on users’ past behavior and profiles. It would be very interesting to explore how commonly used algorithms of a recommender system work through code implementation and compare their performance and time costs.

Data & Tools

• Amazon Review Data: http://snap.stanford.edu/data/web-Amazon-links.html
• Programming language & Libraries: Python 2.7, Numpy, Pandas, Tensorflow
• Database Management System: MySQL
• Computing Instance: AWS EC2 m4.large

Task Assignment:

• Data Preprocessing (CH)
• Data Reader (CH)
• Content-based Recommender System (CH)
• Collaborative Filtering (CH)
• Weight Learned (SM)
• Latent Factor Model (SM)
• Bias Extension (SM)

Dataset Profile

Dataset range	Data size	Min # of reviews in each product	Min # of reviews in each user	# of unique users	# of unique items
Large	1121296	100	10	46824	5801
Small	39126	150	15	1504	165

Instructions

1. Download data

You need to download data to the directory HOME/data. This will take some time.

chmod +x download.sh; ./download.sh 

2. Preprocess data and Create DB

python prepro/preprocess.py

3. Run Recommenders

python -m model.main

It builds a recommender with the training data and also evaluates the performance on the test data. If you want to specify a certain recommender system, you can use '--recom'.

Content Based : 'cb' Collaborative Filtering : 'cf' Weight Learned : 'l' Latent Factor : 'lf' Latent Factor with Bias Extension : 'blf'

For example, if you want to run Weight Learned Recommender,

python -m model.main --recom l

If you want to run both Content Based and Collaborative Filtering Recommenders,

python -m model.main --recom cb cf

It runs a recommender on the small dataset by default. If you want to run on a large dataset, you can use '--small False'. Batch size is 128 by default. If you want to change it, you can use '--batch_size'. For example,

python -m model.main --small False --batch_size 256

References

• [1] G. Adomavicius and A. Tuzhilin. 2005. “Towards the next generation of recommender systems: a survey of the state-of-the-art and possible extensions.” IEEE Transactions on Knowledge and Data Engineering 17 (6): 734– 749.
• [2] Y Koren, R Bell, C Volinsky. 2009. “Matrix factorization techniques for recommender systems.” Computer 42 (8): 30-37.
• [3] J. McAuley, C. Targett, J. Shi, A. van den Hengel. 2015. “Image-based recommendations on styles and substitutes.” SIGIR.