README

Description

This project implements several recommendation system algorithms and an evaluation framework for the patient-narritive dataset. Specifically, Random, UserCF, ItemCF and Pixie[1] are already implemented. The metrics are Mean Reciprocal Rank (MAP), Mean Average Precision (MAP) and Normalized Discount Cummulative Gain (NDCG).

Usage

python main.py

The main.py will run several experiments on the four policies and evaluate the average performance. The result is

$ python main.py
100%|████████████████████████████████████████████████████████████████| 50/50 [00:48<00:00,  1.04s/it]

Method	MRR
Random	0.091
UserCF	0.166
ItemCF	0.075
Pixie	0.171

Method	MAP
Random	0.011
UserCF	0.012
ItemCF	0.009
Pixie	0.009

Method	NDCG
Random	1.175
UserCF	1.378
ItemCF	1.131
Pixie	1.359

[1] Pixie: A system for Recommending 1+ Billion Items to 175+ Million Pinterest Users in Real-Time