AutoRec: Autoencoders Meet Collaborative Filtering

[RytTnk]

I can not remove this comment.

[RytTnk]

Comment

• The first paper on application with AE on Recsys, maybe.
  • He said this is for CF. (I think it ---)
• Number of estimation parameter: 2mk + m + k ???(do calculate)

BG

• MF is non-efficient computation using both item and user interaction to decompose Matrix into each latent reprentitve vecter.

Approach in this work, Model

• Using simple AE but differing from use of visible input and regularization.
• Common recomendation domain(and Scenario) Ml and Netflix
• Item (or user) based-Recommendation

Experiment setting & Evaluation

• Dataset
  • Movielens(1M small, 10M middle size),
  • Netflix()
• As competitive models, BMF, CF-RBM() and LLORMF,
• Metric
  • RMSE
• Tunning
  • Hyperparameter
  • unit_size,
  • hidden layer number,
  • activation function
• Setting
  • Use ONLY 3 STAR (for equal?paradigm)

Result

• Outperformed item-based AE than user-based one as (rich data? less bias?)

Discuusion

• Why? ()

What you should do next

• AE for mechinsm and find improvement (Bailu? 2012? totu, AE)
• VAE tuto, and appied to Rec. -- not gaussian dist.

Ref

• Suvash Sedhain, Aditya Krishna Menon, Scott Sanner, and Lexing Xie. 2015. AutoRec: Autoencoders Meet Collaborative Filtering. In Proceedings of the 24th International Conference on World Wide Web (WWW '15 Companion). ACM, New York, NY, USA, 111-112. DOI: http://dx.doi.org/10.1145/2740908.2742726