MV-HAN: A Hybrid Attentive Networks based Multi-View Learning Model for Large-scale Contents Recommendation

In this paper, we propose a novel model called MV-HAN for the matching stage in recommender systems. We design a hybrid neural structure configured with different models, including MLPs and multi-head self-attentive neural networks. The proposed method transfers the knowledge from the source types to the target types, which helps better representation learning for users and contents. Moreover, the MV-HAN shares parameters of the bottom networks to mitigate the cold start on the spare types. Offline experiment results on industrial datasets show that the proposed method outperforms different baselines, i.e., achieving up to 4.43\% and 4.64\% higher AUC and HR than the best results of baseline methods on the SC dataset. Online experiment results on real-world recommender systems show that the MV-HAN significantly improves the recommendation performance compared with baseline methods in all metrics. It verifies that the MV-HAN is able to handle multi-source asynchronous dataflows and extract information from different content types in real-world applications.

Recommended citation: ‘Ge Fan, Chaoyun Zhang, Kai Wang, and Junyang Chen. 2022. MV-HAN: A Hybrid Attentive Networks based Multi-View Learning Model for Largescale Contents Recommendation. In 37th IEEE/ACM International Conference on Automated Software Engineering (ASE ’22), October 10–14, 2022, Rochester, MI, USA. ACM, New York, NY, USA, 5 pages. https://doi.org/10.1145/3551349.3559496’

@INPROCEEDINGS{fan2022mvhan,
  author={Fan, Ge and Zhang, Chaoyun and Wang, Kai and Chen, Junyang},
  booktitle={37th IEEE/ACM International Conference on Automated Software Engineering (ASE ’22)}, 
  title={MV-HAN: A Hybrid Attentive Networks based Multi-View Learning Model for Largescale Contents Recommendation}, 
  year={2022},
  volume={},
  number={},
  pages={},
  doi={10.1145/3551349.3559496}}

Download paper here