Collaborative Denoising Auto-Encoder for Top-N Recommender Systems

[RytTnk]

(CDAE)

• summarizing some scenarios in
• comprehensive experiments (tuning, competitive )

Comment

• Applying Denoising Auto encoder for Recsys
• Add user specific node to encoder, matrix V tables each u instinct?
• Generalization model of several previous ones
• (common traditional recommendation scenario)

Approach in this work, Model

• Using DAE and L2 regularization.
• User specific node to encoder, matrix V tables each u instinct?

Overview of model-based recommendation

• various model
• loss(point or pair)

Experiment setting & Evaluation

• Dataset

  • Movielens (10M middle size),
  • Netflix()
  • Yelp

• Condition

  • 5-fold cross vali, mean
  • SDG, Ada-grad($beta$=1, $n?$=[1,,,,0,001])
  • probabilistic corruption, rate (For discrete value) [0, 0.2,,,,1.0]
  • NS=5 to z ???

• Pre process, setting

  • no less than 4 star rating, other missing, convert 1 or 0 [ref prev ...]
  • stemming less few than 5 for u, i
  • splitting 80:20

• As competitive models, ,

• Metric

  • P, R, MAP{1,5,10}

• Tunning

  • Hyper parameter
  • activation function, enc , dec

Result

• Outperformed (DAE) as (rich data? less bias?)
• than AutoRec
• Tied Weight
• shown () result than SOA

Discuusion

-

What you should do next

• DAE tuto, and appied to Rec more. For self deep search to improve it with my idea

  • hidden layer number,

Ref

• Yao Wu, Christopher DuBois, Alice X. Zheng, and Martin Ester. 2016. Collaborative Denoising Auto-Encoders for Top-N Recommender Systems. In Proceedings of the Ninth ACM International Conference on Web Search and Data Mining (WSDM '16). ACM, New York, NY, USA, 153-162. DOI: https://doi.org/10.1145/2835776.2835837