In this paper, we propose to implement a piecewise linear model to solve nonlinear classification problems. In order to realize a switch between linear models, a data‐dependent gating mechanism achieved by an autoencoder is designed to assign gate signals automatically. We ensure that a diversity of gate signals is available so that it is possible for our model to switch between a large number of linear classifiers. Besides, we also introduce a sparsity level to add a manual control on the flexibility of the proposed model by using a winner‐take‐all strategy. Therefore, our model can maintain a balance between underfitting and overfitting problems. Then, given a learned gating mechanism, the proposed model is shown to be equivalent to a kernel machine by deriving a quasilinear kernel function with the gating mechanism included. Therefore, a quasilinear support vector machine can be applied to solve the nonlinear classification problems. Experimental results demonstrate that our proposed piecewise linear model performs better than or is at least competitive with its state‐of‐the‐art counterparts. © 2019 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

With the rapid development and extensive application of Web services, various approaches for Web service recommendation have been proposed in the past. However, the traditional methods only utilize the information of the user-service rating matrix but ignore the trust relations between users, so their recommendation precision is often unsatisfactory, and, furthermore, most of these methods lack the ability to distinguish the credibility of recommendation. To address the problems, we proposed a personalized service recommendation based on trust relationship. In particular, our approach takes into account user experience, interest background, recommendation effect, and evaluation tendency in the formalization of trust relationship, and moreover it can filter out useless or suspected services by exploiting trust relationships between users. To verify the proposed approach, we conducted experiments by using a real-world Web services set. The experimental results show that our proposed approach leads to a substantial increase in the precision and the credibility of service recommendations.

Abstract:In order to alleviate the pressure of information overload and enhance consumer satisfaction, personalization recommendation has become increasingly popular in recent years. As a result, various approaches for recommendation have been proposed in the past few years. However, traditional recommendation methods are still troubled with typical issues such as cold start, sparsity, and low accuracy. To address these problems, this paper proposed an improved recommendation method based on trust relationships in social networks to improve the performance of recommendations. In particular, we define trust relationship afresh and consider several representative factors in the formalization of trust relationships. To verify the proposed approach comprehensively, this paper conducted experiments in three ways. The experimental results show that our proposed approach leads to a substantial increase in prediction accuracy and is very helpful in dealing with cold start and sparsity.

No abstract

This paper proposes a novel oversampling method for imbalanced data classification, in which the minority class samples are synthesized in a feature space to avoid the generated minority samples falling into the majority class regions. For this purpose, it introduces a multi-linear feature space (MLFS) based on a quasi-linear kernel, which is composed from a pretrained neural network (NN). By using the quasi-linear kernel, the proposed MLFS oversampling method avoids computing directly the Euclidean distances among the samples when oversampling the minority class and mapping the samples to high-dimensional feature space, which makes it easy to be applied to classification of high-dimensional datasets. On the other hand, by using kernel learning instead of representation learning using the NN, it makes an unsupervised learning, even a transfer learning, to be easily employed for the pretraining of NNs because a kernel is usually less dependent on a specific problem, which makes it possible to avoid considering the imbalance problem at the stage of pretraining the NN. Finally, a method is developed to oversample the synthetic minority samples by computing the quasi-linear kernel matrix instead of computing very high dimensional MLFS feature vectors directly. The proposed MLFS oversampling method is applied to different real-world datasets including image dataset, and simulation results confirm the effectiveness of the proposed method.