Data Cleaning by Deep Dictionary Learning

Zhongqi Lu, Qiang Yang

Feb 18, 2016 (modified: Feb 18, 2016) ICLR 2016 workshop submission readers: everyone
  • Abstract: The soundness of training data is important to the performance of a learning model. However in recommender systems, the training data are usually noisy, because of the randomness nature of users' behaviors and the sparseness of the users' feedback towards the recommendations. In this work, we would like to propose a noise elimination model to preprocess the training data in recommender systems. We define the noise as the abnormal patterns in the users' feedback. The proposed deep dictionary learning model tries to find the common patterns through dictionary learning. We define a dictionary through the output layer of a stacked autoencoder, so that the dictionary is represented by a deep structure and the noise in the dictionary is further filtered-out.
  • Conflicts: ust.hk, cse.ust.hk

Loading