whonor / MINS

Multi-Interest News Sequence model

MINS

The implement of Multi-Interest News Sequence model.

Wang, R., Wang, S., Lu, W., Peng, X.: News recommendation via multi-interest
news sequence modelling. In: ICASSP. pp. 7942–7946 (2022)

Requirement

• python~=3.8
• pytorch~=1.9.0
• numpy~=1.20.1
• pandas~=1.2.4
• tensorboard~=2.6
• tqdm~=4.59.0
• nltk~=3.6.2
• scikit-learn~=0.24.1

Dataset

# Download GloVe pre-trained word embedding
https://nlp.stanford.edu/data/glove.840B.300d.zip

# Download MIND dataset
https://msnews.github.io/.

Run

# Train the model, meanwhile save checkpoints
python3 src/train1.py
# Load latest checkpoint and evaluate on the test set
python3 src/evaluate.py

Acknowledgement

Any scientific publications that use our codes and datasets should cite the following paper as the reference:

@inproceedings{wang2022news,
  title={News recommendation via multi-interest news sequence modelling},
  author={Wang, Rongyao and Wang, Shoujin and Lu, Wenpeng and Peng, Xueping},
  booktitle={ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={7942--7946},
  year={2022},
  organization={IEEE}
}

Credits

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