whonor / NewsRecommendation

Implementations of some methods in news recommendation.

News Recommendation

The repository currently includes the following models.

Model	Full name	Paper
NRMS	Neural News Recommendation with Multi-Head Self-Attention	https://www.aclweb.org/anthology/D19-1671/
NAML	Neural News Recommendation with Attentive Multi-View Learning	https://arxiv.org/abs/1907.05576
LSTUR	Neural News Recommendation with Long- and Short-term User Representations	https://www.aclweb.org/anthology/P19-1033.pdf
DKN	Deep Knowledge-Aware Network for News Recommendation	https://dl.acm.org/doi/abs/10.1145/3178876.3186175
Hi-Fi Ark	Deep User Representation via High-Fidelity Archive Network	https://www.ijcai.org/Proceedings/2019/424
TANR	Neural News Recommendation with Topic-Aware News Representation	https://www.aclweb.org/anthology/P19-1110.pdf
FIM	Fine-grained Interest Matching for Neural News Recommendation

Get started

Basic setup.

git clone https://github.com/yusanshi/NewsRecommendation
cd NewsRecommendation
pip3 install -r requirements.txt

Download and preprocess the data.

mkdir data && cd data
# Download GloVe pre-trained word embedding
wget https://nlp.stanford.edu/data/glove.840B.300d.zip
sudo apt install unzip
unzip glove.840B.300d.zip -d glove
rm glove.840B.300d.zip

# Download MIND dataset
# By downloading the dataset, you agree to the [Microsoft Research License Terms](https://go.microsoft.com/fwlink/?LinkID=206977). For more detail about the dataset, see https://msnews.github.io/.
wget https://mind201910small.blob.core.windows.net/release/MINDlarge_train.zip https://mind201910small.blob.core.windows.net/release/MINDlarge_dev.zip
unzip MINDlarge_train.zip -d train
unzip MINDlarge_dev.zip -d test
rm MINDlarge_*.zip

# Preprocess data into appropriate format
cd ..
python3 src/data_preprocess.py
# Remember you shoud modify `num_*` in `src/config.py` by the output of `src/data_preprocess.py`

Modify src/config.py to select target model. The configuration file is organized into general part (which is applied to all models) and model-specific part (that some models not have).

vim src/config.py

Run.

# Train and save checkpoint into `checkpoint/{model_name}/` directory
python3 src/train.py
# Load latest checkpoint and evaluate on the test set
# This will also generate answer file `data/test/answer.json`
python3 src/evaluate.py

# or

chmod +x run.sh
./run.sh

You can visualize metrics with TensorBoard.

tensorboard --logdir=runs

# or
tensorboard --logdir=runs/{model_name}
# for a specific model

Tip: by adding REMARK environment variable, you can make the runs name in TensorBoard more meaningful. For example, REMARK=num-filters-300-window-size-5 python3 src/train.py.

Results

Update: the table is out-of-date.

Model	AUC	nMRR	nDCG@5	nDCG@10	Remark
NRMS	65.49	30.28	33.07	39.62	ckpt-3200.pth, batch size 192
NAML	66.63	31.72	35.00	41.15	ckpt-8000.pth, batch size 64
LSTUR	62.67	28.21	30.51	37.15	ckpt-8800.pth, batch size 128
DKN	62.35	28.09	30.56	37.16	ckpt-11400.pth, batch size 64
Hi-Fi Ark	64.15	29.66	32.35	38.85	ckpt-5000.pth, batch size 128
TANR	66.55	31.62	34.96	41.12	ckpt-3800.pth, batch size 128
FIM

Checkpoints: https://drive.google.com/open?id=1GbhxYd2qifRImFK4b883x7xXU-lpoJx-

You can verify the results by simply downloading them and running MODEL_NAME=XXXX python3 src/evaluate.py.

Credits

• Dataset by MIcrosoft News Dataset (MIND), see https://msnews.github.io/.