Research Of Parallelized Collaborative Deeplearning

Collaborative Deep Learning relieves the problem that Conventional CF-based methods use the ratings given to items by users as the sole source of information for learning to make recommendation,while CF-based methods to degrade significantly in their recommendation performance because of the ratings are very sparse.CDL use the ability of Neural network that learn the features from data and to fit a robust model automatically.It introduces a new way to solve the problems that the recommendation system meets the very sparse data.Meanwhile,Recommendation System can not hold the Big Data because of the Rating Matrix is too large to load in the memory and the process of training for the content model is hard to predict.As a result,the Distribute Platform for RS is more and more necessary.To solve the problem mentioned above,We propose in this paper a model based on CDL called "CDL with Item private Node"(CDL-i),And train CDL-i with a parallel method on Apache Spark.When training the CDL-i model,more than one dataset was used.we transfer CDL-i to Spark to investigate it's scalability.First of this paper,we introduce the theory of the Parallelized CDL-i,including CDL,Autoencoder,PMF,state of parallelized machine learning and Apache Spark.Then,we develop a modified collaborative deep learning CDL-i.Meanwhile,we introduce a method to show that how to transfer CDL-i to Spark.By the way,a framework based Spark for training machine learning model with pipeline Logic was developed.At last,extensive experiments on real-world datasets show that our modified CDL-i can improve the performance of original CDL.And Parallel experiments can tell us the parallelized CDL was scalable and effective.As the result,CDL-i can perform a better precision,and has the ability to hold the data which has a huge size.The framework helps the algorithm easy to use.