TRAP

TRAP: Two-level Regularized Autoencoder-based Embedding for Power-law Distributed Data

TRAP is a general and powerful regularizer for autoencoder (AE)-based embedding methods on the graph data where the data follow the power-law distribution w.r.t the sparsity of input vectors. TRAP significantly boosts performances of two represensitive graph embedding tasks, (1) Top-k recommendation on user-item transaction datasets and (2) Node classification on common graph datasets, by up to 31.53% and 94.99% respectively.

Citations

@inproceedings{park2020trap,
  title={TRAP: Two-level regularized autoencoder-based embedding for power-law distributed data},
  author={Park, Dongmin and Song, Hwanjun and Kim, Minseok and Lee, Jae-Gil},
  booktitle={Proceedings of The Web Conference 2020},
  pages={1615--1624},
  year={2020}
}

Paper/Video Links

[paper] [video]

Setup

• python3
• tensorflow v1.14.0

Usage

Two tasks are used to evaluate the effectiveness of TRAP; (1)Top-k recommendation with user-item embedding and (2) Node classification with node embedding. The source code of each task is in the UserEmbedding and NodeEmbedding folders, respectively. Detail instructions of how to run the codes can be found on Readme.md files in each folder.

Code References

Because TRAP is a meta-algorithm (regularizer), we combined it with existing baseline methods to improve the performance. The github baseline codes we mainly used are the following:

User-item Embedding

• https://github.com/Zziwei/Joint-Collaborative-Autoencoder, Joint Collaborative Autoencoder (JCA), TheWebConf, 2019.

Node Embedding

• https://github.com/suanrong/SDNE, Structural Deep Network embedding (SDNE), SIGKDD, 2016.

Thank you for the authors.

Contact

Please post a Github issue or contact dongminpark@kaist.ac.kr if you have any questions. Thanks.