Search

Software comprehension helps us to maintain software better. The developers often use documentation and mutual communication to understand software. However, the incompletion of documentation usually causes some problems such that we cannot acquire the detailed information of software. So researchers are now using key classes as starting points for software understanding and propose several effective methods for key class identification. Such methods are mostly implemented using supervised and unsupervised learning methods; however, they may be subject to class imbalance, insufficient amount of data, and lack of accuracy. In this paper, we propose a novel approach, called COSPA (identifying key Classes in Object-Oriented Software using Preference Aggregation), to identify key classes in software. Our approach is inspired by the social choice theory and uses Kemeny-Young method to find a preference that minimizes the total difference between different pairs of sequences returned by three key class identification methods (i.e., a-index, InDeg, and OSE). Further, we rank the classes in descending order according to that preference set, and a small number of top-ranked classes are considered as key candidates. To better evaluate the performance of our approach, we apply it to six Java software systems and compare it with nine state-of-the-art approaches. In the experiments, we use Friedman to evaluate all methods in the whole set of research subjects; results show that, though no single method works always best in all subject systems, our method indeed performs better than others.

Cloud applications based on service-oriented architectures usually integrate many component services to implement specific application logic. In service-oriented computing environments, many Web services are provided for users to build service-oriented systems. Since the performance of the same Web service varies according to different users' perspectives, the users have to personally select the optimal Web services according to the quality-of-service (QoS) data observed by other similar users. However, users with a low reputation provide unreliable data, which has a negative impact on service selection. Moreover, the QoS data vary over time due to changes in user reputation; and therefore, how to calculate a personalized reputation for each user at runtime remains a substantial problem. To address this critical challenge, this paper proposes an online reputation calculation method, called the OPRC, to efficiently provide a personalized reputation for each user. Based on the users' observed QoS data, the OPRC employs MF and online learning techniques to calculate personalized reputations. To validate the approach, large-scale experiments are conducted, which contain two QoS attributes from 142 reliable users and 15 unreliable users. The results show that OPRC has high accuracy and effectiveness compared to other approaches.

To improve the performance of network intrusion detection systems (IDS), we applied deep learning theory to intrusion detection and developed a deep network model with automatic feature extraction. In this paper, we consider the characteristics of the time-related intrusion and propose a novel IDS that consists of a recurrent neural network with gated recurrent units (GRU), multilayer perceptron (MLP), and softmax module. Experiments on the well-known KDD 99 and NSL-KDD data sets show that the system has leading performance. The overall detection rate was 99.42% using KDD 99 and 99.31% using NSL-KDD with false positive rates as low as 0.05% and 0.84%, respectively. In particular, for detecting the denial of service attacks, the system achieved detection rates of 99.98% and 99.55%, respectively. Comparative experiments showed that the GRU is more suitable as a memory unit for IDS than LSTM, and proved that it is an effective simplification and improvement of LSTM. Moreover, the bidirectional GRU can reach the best performance compared with the recently published methods.

Over the last decade, many pairwise-constraint-based metric learning algorithms have been developed to automatically learn application-specific metrics from data under similarity/dissimilarity data-pair constraints (weak labels). Nevertheless, these existing methods are designed for the centralized learning case, in which all the data and constraints are supposed to be gathered together in one source, and the algorithms utilize the whole data and constraints information during the learning process. However, in many real applications, large amounts of data (constraints) are dispersedly generated/stored in geographically distributed nodes over networks. Thus, it might be impractical to centralize the whole data information to one fusion node. Besides, in such cases, it is often hard to have every data pair labeled due to the huge data-pair amounts, resulting in numerous unlabeled data pairs. Given these situations, in this paper, we propose two types, namely, a diffusion type and an alternating-direction-method-of-multipliers type, of distributed semi-supervised metric learning frameworks, which make use of both labeled and unlabeled data pairs. The proposed frameworks can be easily used to extend centralized metric learning methods of different objective functions to distributed cases. In particular, we apply our frameworks on a well-behaved centralized semi-supervised metric learning method called SERAPH and yield two new distributed semi-supervised metric learning algorithms. Our simulation results show that the metrics learned by the proposed distributed algorithms are very close to that of the corresponding centralized method in most cases.