改进的多层感知机在客户流失预测中的应用

针对在传统的客户流失预测数据预处理中，使用one-hot编码处理离散属性导致数据维度增加及数据过于稀疏的问题，提出了两种基于多层感知机的改进后的客户流失预测模型。其主要思想是分别使用堆叠自编码器和实体嵌入两种方法对多层感知机进行改进，通过将离散属性的高维编码数据向低维空间映射，有效地减少了one-hot编码产生的稀疏数据，增加了离散属性值之间的关联度。在对两份公开的数据集进行交叉验证后的实验结果表明，改进后的模型既有效地提高了预测的准确度，又维持了传统多层感知机模型在并行化计算方面的优势。     

A participatory,farmer-led approach to changing practices around antimicrobial use on UK farms

Farmer-led, participatory approaches are being increasingly employed in agricultural research, with promising results. This study aimed to understand how a participatory approach based on the Danish stable schools could help to achieve practical, farmer-led changes that reduced reliance on antimicrobials in the UK. Five facilitated farmer action groups comprising 30 dairy farms across South West England met on farm at regular intervals between 2016 and 2018, and worked collaboratively within their groups to discuss how to reduce antimicrobial use. Qualitative data from group discussions and individual semi-structured interviews were collected and analyzed using thematic analysis to explore how the approach helped farmers address and deal with changes to their on-farm practices. Facilitator-guided reviews of antimicrobial use and benchmarking were carried out on each farm to assess any change in usage and help farmers review their practices. The pattern of antimicrobial use changed over the 2 yr of the study, with 21 participating farms reducing their use of highest-priority critically important antibiotics (6 farms were not using any of these critical medicines from the outset). Thirty practical action plans were co-developed by the groups with an average implementation rate of 54.3% within a year. All assessed farms implemented 1 recommendation, and many were still ongoing at the end of the study. Farmers particularly valued the peer-to-peer learning during farm walks. Farmers reported how facilitated discussions and action planning as a peer group had empowered them to change practices. Participants identified knowledge gaps during the project, particularly on highest-priority critically important antibiotics, where they were not getting information from their veterinarians. The study demonstrated that facilitation has a valuable role to play in participatory approaches beyond moderating discussion; facilitators encouraged knowledge mobilization within the groups and were participants in the research as well. Facilitated, farmer-led, participatory approaches that mobilize different forms of knowledge and encourage peer learning are a promising way of helping farmers to adapt and develop responsible practices.     

Key-Attributes-Based Ensemble Classifier for Customer Churn Prediction

Recently, it has been seen that the ensemble classifier is an effective way to enhance the prediction performance. However, it usually suffers from the problem of how to construct an appropriate classifier based on a set of complex data, for example, the data with many dimensions or hierarchical attributes. This study proposes a method to constructe an ensemble classifier based on the key attributes. In addition to its high-performance on precision shared by common ensemble classifiers, the calculation results are highly intelligible and thus easy for understanding. Furthermore, the experimental results based on the real data collected from China Mobile show that the key-attributes-based ensemble classifier has the good performance on both of the classifier construction and the customer churn prediction.     

A survey on privacy in decentralized online social networks

Decentralized Online Social Networks (DOSNs) have recently captured the interest of users because of the more control given to them over their shared contents. Indeed, most of the user privacy issues related to the centralized Online Social Network (OSN) services (such as Facebook or Google+) do not apply in the case of DOSNs because of the absence of the centralized service provider. However, these new architectures have motivated researchers to investigate new privacy solutions that allow DOSN’s users to protect their contents by taking into account the decentralized nature of the DOSNs platform.In this survey, we provide a comprehensive overview of the privacy solutions adopted by currently available DOSNs, and we compare them by exploiting several criteria. After presenting the differences that existing DOSNs present in terms of provided services and architecture, we identify, for each of them, the privacy model used to define the privacy policies and the mechanisms for their management (i.e., initialization and modification of the privacy policy). In addition, we evaluate the overhead introduced by the security mechanisms adopted for privacy policy management and enforcement by discussing their advantages and drawbacks.     

Revisiting Linus’s law: Benefits and challenges of open source software peer review

Open source projects leverage a large number of people to review products and improve code quality. Differences among participants are inevitable and important to this collaborative review process—participants with different expertise, experience, resources, and values approach the problems differently, increasing the likelihood of finding more bugs and fixing the particularly difficult ones. To understand the impacts of member differences on the open source software peer review process, we examined bug reports of Mozilla Firefox. These analyses show that the various types of member differences increase workload as well as frustration and conflicts. However, they facilitate situated learning, problem characterization, design review, and boundary spanning. We discuss implications for work performance and community engagement, and suggest several ways to leverage member differences in the open source software peer review process.     

Adolescents’ comments in social media: Why do adolescents receive negative feedback and who is most at risk?

Receiving negative peer feedback in social media may have negative consequences for adolescents’ psychosocial development and well-being. Therefore, the first aim of this study was to investigate online behavior (i.e., online social exploration, risky online self-presentation) that predicts receiving negative online peer feedback. The second aim was to examine three types of precursors that may predict this online behavior and, indirectly, negative feedback: (a) developmental (i.e., sex, age), (b) dispositional (i.e., sensation seeking, inhibitory control), and (c) social precursors (i.e., peer problems, family conflict). We collected survey data among 785 Dutch adolescents (10–15 years old). Our results showed that adolescents who engaged in online social exploration and risky online self-presentation more often, were more likely to receive negative peer feedback. Online social exploration was more prevalent among the older adolescents and adolescents characterized by higher sensation seeking and more family conflict. In addition, risky online self-presentation was more prevalent among adolescents high in sensation seeking. Consequently, these adolescents’ online behavior, indirectly, made them more at risk of receiving negative peer feedback in social media.     

Top-k query processing for replicated data in mobile peer to peer networks

In mobile ad hoc peer to peer (M-P2P) networks, since nodes are highly resource constrained, it is effective to retrieve data items using a top-k query, in which data items are ordered by the score of a particular attribute and the query-issuing node acquires data items with the k highest scores. However, when network partitioning occurs, the query-issuing node cannot connect to some nodes having data items included in the top-k query result, and thus, the accuracy of the query result decreases. To solve this problem, data replication is a promising approach. However, if each node sends back its own data items (replicas) responding to a query without considering replicas held by others, same data items are sent back to the query-issuing node more than once through long paths, which results in increase of traffic. In this paper, we propose a top-k query processing method considering data replication in M-P2P networks. This method suppresses duplicate transmissions of same data items through long paths. Moreover, an intermediate node stops transmitting a query message on-demand.     

Software defined P2P architecture for reliable vehicular communications

Vehicular networking is considered a promising and enabling technology that helps in the realization and diversification of vehicle related applications such as road traffic safety, emergency management, and infotainment. For applications such as the road traffic safety, the system has to respond in certain bounded time and ensure higher reliability. A major bottleneck of the existing architectures is the reliability and scalability that results in a considerable performance degradation. Therefore, in this paper, we proposed a software defined peer to peer (P2P) architecture for reliable vehicular communications solution that scales out overall network intelligence into the system components (Road side units, on-board units, Gateways, Control centre) and offer an unprecedented reliability, adaptability and scalability. The reliability of the communication is achieved by using redundant system components such as multiple switches and GWs links that are deployed to achieve lower communication latencies and higher packet success rates. Furthermore, higher reliability is achieved by the application of fast-failover and fast recovery techniques. The proposed architecture is implemented using Docker, and containerized ONOS and Cassandra. The obtained results show that the proposed architecture offer higher reliability by offering higher packet success rate (PSR) and smaller round trip time (RTT).     

BitTorrent系统节点行为研究

测量分析P2P文件共享系统是解决P2P优化、网络管理等问题的基础。当前P2P网络设计日趋复杂,并且部分运行机制直接导致测量数据不完整或测量周期过长,比如BitTorrent文件共享系统。针对这一问题,本文抽象出BitTorrent系统的3个测量层次,包括索引测量、节点测量和内容测量,并重点解决了3个层次中的测量难题,设计实现了一个面向BitTorrent的测量系统。在此基础上,本文对BitTorrent系统进行了持续测量,并重点分析了BitTorrent系统的节点行为:节点演化规律和节点到达模型。测量结果表明,节点演化规律和节点到达均呈现强烈的周期性。为此,本文改进了节点到达模型,新模型与实际测量数据拟合效果较好,进一步修正了节点到达模型。