期刊名称：IAENG International Journal of Computer Science
印刷版ISSN：1819-656X
电子版ISSN：1819-9224
出版年度：2019
卷号：46
期号：1
页码：109-117
出版社：IAENG - International Association of Engineers
摘要：Deep learning has demonstrated some remarkablesuccesses. Deep learning can be characterized in several differentways, but the most important is that deep learning canlearn higher-order interactions among features using a cascadeof many layers. Despite its successes, it faces major challengesin computational complexity when using hyper-parameters andtime-consuming processes to calculate many fully connectedlayers. We propose an alternate approach to facilitate deeplearning with imbalanced datasets. We call our model ImbalanceDeep Belief Network (IDBN). We incorporate ensemblemodels to enable the system to learn all the class targets (userrating) with a single deep learning base classifier model basedon user comments. By this way, it can learn every rating withits fitting model. The main idea for the ensemble model is togive an alternate solution for base classifiers when selecting thebest results. In the ensemble model, we use different featuresselection for the input layer. Using all features in the same modelmay increase computational complexity and consume muchtime. Further, some features can be assessed for significance.Insignificant features could be pruned in the classifier or besubstituted with other features which are extracted from thesampling data. In the output layer, we apply voting methodsto select user ratings as outputs from each base classifier togenerate user rating prediction results with more extensiveresults. IDBN has more sustainably predicts bad and very bad’user ratings in imbalanced datasets than base models.
关键词：Deep Learning; User Rating Prediction; Imbalance;Classification
