DIAN: A TensorFlow Library for Implicit Data Recommendation

Installation

git clone https://github.com/lzheng21/DIAN.git
cd DIAN
python setup.py install

Supported Models

Bayesian Personalized Ranking (https://arxiv.org/abs/1205.2618)

Factorization Machines (https://www.csie.ntu.edu.tw/~b97053/paper/Rendle2010FM.pdf)

Dependencies

Python 3.5
TensorFlow 1.5.0
NumPy >= 1.4.0

Input Data Format

train file:

uid1 iid_1 iid_2 ...
uid2 iid_5 iid_3 ...

test file:

uid1 iid_3 iid_4 ...
uid2 iid_7 iid_8 ...

Usage

from DIAN.data import Data
from DIAN.models.BPR import BPR
data = Data(train_file='data/ml-1m/'+'train_users.dat', test_file='data/ml-1m/'+'test_users.dat')
model = BPR(data, emb_dim=16, batch_size=1024)
model.train(200, lr=0.01, optimizer='Adam')
model.predict()

Performance

Model   HitRatio@20 MAP@20
BPR     0.20        0.06
FM      0.18        0.04