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。     

Osmotic dehydration pretreatment for improving the quality attributes of frozen mango: effects of different osmotic solutes and concentrations on the samples

The effect of osmotic dehydration pretreatment on the quality attributes (e.g. colour, hardness, drip loss, vitamin C content and pH) of frozen mango cuboids in different osmotic solutions (sucrose, glucose and maltose) and concentrations (30, 45 and 60% (wt/wt)) was investigated. Results revealed that melting temperature of mango cuboids was affected by both solute type and solid mass fraction. In addition, the dehydrofrozen samples pretreated in maltose had higher quality in vitamin C content (increasing by 23.5–73.0%), colour (colour change reducing by 2.6–39.2%) and drip loss (reducing by 0.7–9.7%) than those pretreated in other osmotic solutions. The cuboids pretreated in glucose displayed higher hardness (increasing by 16.4–36.2%). Based on principal component analysis and group distance, osmotic dehydration in 45% maltose was proposed as the most favourable freezing conditions with the highest sensory score (6.8). The current work indicates osmotic dehydration significantly improves frozen mango quality.     

Evaluation across and within collaborative manufacturing networks: a comparison of manufacturers’ interactions and attributes

Evaluation on collaborative manufacturing network (CMN) structure characteristics has important implications for network operations such as production decision-making, product recovery, creating consensus. Several recent studies suggest that augmenting network structure with nodes’ attributes can provide a more fine-grained understanding of the network. However, there have been few studies to provide a systematic understanding of these effects in a CMN at scale. This gap is bridged using an industrial printing machinery CMN data-set collected on a web-based producing and outsourcing service platform. Novel phenomena with respect to both interaction and attribute metrics across and within the CMNs are observed. Moreover, although many studies employ either interaction or attribute data to study the relative roles of manufacturers in a CMN, relatively little is known about the relationship between these two types of data. This study explores this relationship by comparing two defined metrics (i.e. relational capability and node capability), which evaluate the manufacturers’ interactions and attributes, respectively. We examine to what extent the two metrics of manufacturers correlate with each other, and how possible dissimilarities and similarities can be explained based on the collected industrial CMN data-set. The insights thereby provide a better basis for efficient operations decision-making in CMN.     

An exhaustive criterion for estimating quality of images in electrical impedance tomography with application to clinical imaging

This study proposes a versatile criterion for estimating quality of images in electrical impedance tomography. The point spread function (PSF) is calculated throughout the domain based on the scattering of energy as responses to a small anomaly spirally moved from the centre to the boundary. The proposed PSF is a measure of weighted spatial variance (WSV) of the conductivity over the whole domain. For each element, the weighting factor is a normalized multiplication of the area of that element by its square intensity. The WSV collectively incorporates all image attributes, i.e., spatial resolution, artifact, amplitude response, positioning error and shape deformation. The location of artifacts, which significantly influences reconstructed images in reality, is taken into account as well. The results illustrate that the proposed measure is more tolerant than existing criteria in evaluating performance of EIT systems in both theory and practice.     

面向近邻泄露的数值型敏感属性隐私保护方法

提出一种面向近邻泄露的数值型敏感属性隐私保护方法,该方法首先在保护准标识符属性和数值型敏感属性内在关系的前提下,将数值型敏感属性进行离散化划分;然后,提出一种面向近邻泄露的隐私保护原则——(k, ε)- proximity;最后,设计了最大邻域优先算法MNF(maximal neighborhood first)来实现该原则。实验结果表明,提出的方法能在有效保护数值型敏感信息不泄露的同时保持较高的数据效用,并且保护了数据间的关系。     

Project competition intensity (PCI) in the construction market: a case study in China

In economic terminology, ‘competition intensity’ can be used to mirror to a certain degree the status of industrial competition in a specific market, while in the construction market, it particularly involves the part of project competition intensity (PCI). Based on a road project in China, two forms of PCI, overt and covert, are identified in this study. To lower project construction cost thus enabling better project tendering performance, clients can facilitate higher levels of overt PCI by tailoring their tendering procedures, whereas, through covert PCI project competitors can adopt anti‐competitive behaviours to attain the expected works. The difference between overt PCI and covert PCI indicates various practitioners' perspectives and impact on the levels of competition intensity in construction. The findings suggest a way to promote PCI studies in the future, and offer new insights into other parts of competition intensity in construction.     

An evaluation tool for design quality: PFI sheltered housing

The complex procurement process entailed by the private finance initiative (PFI) means that clients need new capacities to manage their relationships with bidders and to assess project proposals if the desired level of design quality is to be achieved. To assist local authorities in their client role, a new Architectural Design Quality Evaluation Tool was developed. The aim was to improve the quality of design in residential sheltered housing, procured through the PFI. The tool was developed for and applied to a programme that will see the replacement of a local authority's entire sheltered housing stock. The tool has two functions: (1) to inform the client's assessment process and assist with the selection of the preferred bidding consortium through a series of stages in the PFI process; and (2) to improve the quality of all the submitted designs through an iterative process. Although several existing mechanisms are available for evaluating the performance attributes of buildings, few also tackle the less tangible amenity attributes, which are vital to the feeling of home. The new tool emphasizes the amenity attributes without neglecting performance.     

一种基于改进差别矩阵的核增量式更新算法

提出一种基于改进差别矩阵的核增量式更新算法,主要考虑对象动态增加情况下核的更新问题．该算法在更新差别矩阵时仅须插入某一行及某一列,或删除某一行并修改相应的列,因而可有效地提高核的更新效率．理论分析和实验结果表明,该算法是有效可行的．     

基于改进的RS-TOPSIS模型的岩爆倾向性预测

针对目前岩爆倾向性中预测模型权重确定存在不足导致模型精度不高的现状,为更准确地预测岩爆倾向性,提出综合运用粗糙集理论中的代数观和信息观,确定属性最优权重,并修正岩爆倾向性与评价指标之间的关系,建立岩爆等级理想点矩阵。根据岩爆发生条件,选取岩石脆性指数、切应力指标和弹性应变能指数3项指标作为岩爆判别指标,以国内外20组典型岩爆数据为样本,建立改进的粗糙集—理想点法（RS-TOPSIS）岩爆倾向性预测模型,并应用该模型对玲珑金矿等工程实际进行了岩爆倾向性预测。结果表明：改进后样本预测精度相比于改进前有了显著提高,所建立的模型对实际工程的岩爆倾向性预测效果良好,预测结果更准确。     

Impact of maceration time on colour‐related phenolics,sensory characteristics and volatile composition of mulberry wine

This study investigated the effect of maceration treatment on colour‐related phenolics, sensory characteristics and volatile composition in mulberry wine. Pre‐fermentation maceration was performed for 2 days (‘short’), 6 days (‘medium’) and 8 days (‘long’), respectively, followed by the same fermentation process. Results showed that the long maceration (LM) caused higher level of anthocyanins in wine, especially cyaniding‐3‐O‐rutinoside, cyaniding‐3‐O‐glucoside and dephinidin‐3‐O‐glucoside, whereas the short maceration caused lower level of phenolic acids in wine. The medium maceration (MM) wine showed the highest level of flavonols but the lowest concentration of flavan‐3‐ol. The LM wine exhibited greater red tone and colour intensity accompanied by higher amounts of free and copigmented anthocyanins. The LM wine showed lower levels of alcohols, acids, terpenoids, benzenes and sulphur compounds, whereas the lowest level of esters was found in the MM wine. The LM wine had reduced fruity and floral aroma and overall flavour complexity compared with the other wines. Copyright © 2018 The Institute of Brewing & Distilling