一种多元核Logistic回归说话人辨别方法

针对文本无关话者辨别多分类目标和大训练样本情况,将经典Logistic回归模型进行多元化变形,并叠加L2惩罚因子以提高模型泛化能力.将最优目标负对数Logistic公式对偶化,并利用序列最小优化算法进行模型训练,速率优于传统多元核Logistic回归训练算法.实验显示,该模型构建简单,训练算法快捷,且识别率优于经典支持向量机与二元核Logistic回归模型所生成的"一对一"多分类方法.     

Logistic回归模型分析

对Logistic回归模型做了比较详细的分析.通过阐述回归分析与概率假定的关系,并把回归模型纳入广义线性模型框架进行推导和分析,便于全面了解回归模型及其理论依据和构造方法,以利于对回归模型的合理应用.     

Hessian正则化Logistic回归模型

针对大数据量的图像分类问题,Laplacian正则化的半监督学习方法获得了广阔的应用前景。然而Laplacian正则化使分类函数趋向于常数函数而易导致较差的推测能力。提出了基于Hessian正则化的Logistic回归模型用于图像分类,Hessian正则化可以较好地预测区域之外的数据点。在MIR Flickr数据库上进行图像分类实验,与SVM、Logistic回归和Laplacian正则化的Logistic回归方法相比,Hessian正则化的Logistic回归模型更有效。     

基于特征矢量集的核Logistic回归

将经典Logistic回归推广到高维RKHS空间，提出了一种基于特征矢量选择的核Logistic回归算法-FVS-KLR．该算法利用特征矢量选择（FVS）从训练样本集中选择一个特征矢量集，原样本在RKHS空间中的映射可用该集合中元素映射的线性组合近似．以特征矢量集为基得到核Logistic回归的目标泛函，并采用Newton-Raphson方法寻优，将优化的计算量由O（N^3）降到O（NL^2），L〈〈N．同时文章推导了多类情况下的核Logistic回归算法．通过与SVM的对比实验表明，该算法对后验概率的估计优于SVM方法，同时在分类错误率不高于SVM的基础上能显著降低分类器的计算量．     

快速成型制造中零件制作方向的优化方法

通过分析STL模型的几何特性和快速成型的工艺过程，建立了台阶效应、支撑面积和制作时间3个目标的优化模型，并分别利用遗传算法进行了求解，建立了独立满意度；然后采用线性加权法建立了综合满意度函数。利用遗传算法求解，得到优化的制作方向．实验结果表明，该方法能够有效地解决快速成型中的制作方向优化问题．     

使用Logistic回归模型进行中文文本分类

使用Logistic回归模型进行中文文本分类,通过实验,比较和分析了不同的中文文本特征、不同的特征数目、不同文档集合的情况下,基于Logistic回归模型的分类器的性能。并将其与线性SVM文本分类器进行了比较,结果显示它的分类性能与线性SVM方法相当,表明这种方法应用于文本分类的有效性。     

基于Logistic回归模型的藏文文本分类研究与实现

文本分类是信息处理领域的核心研究内容,在自动检索和文本过滤等研究领域中被广泛使用。本次研究主要是基于Logistic回归模型分类器对藏文文本进行分类,其核心思想是首先对藏文语料进行收集和预处理,且利用信息增益算法和欧式距离分别对文本特征进行选择与提取;其次构造Logistic回归模型分类器;最后测试和分析分类的准确率、召回率和F1值,同时,对Logistic算法和Gaussian NB算法进行分类性能对比,结果显示Logistic算法具有较好的分类效果。     

纳税人诚信识别系统设计的方法与模型研究

在纳税人税收不遵从行为的微观经济解释基础上,应用Logistic回归模型来识别纳税人诚信,给出纳税人诚信的测算方法。并应用实际数据作实证研究,对模型进行参数估计和预测分析,证实了Logistic回归模型是完善纳税信誉等级管理的强有力技术手段,为设计纳税人诚信识别系统提供了完善的技术解决方案。     

L1/2正则化Logistic回归

提出一种L1/2正则化Logistic回归模型,并针对此模型构造有效的求解算法.文中模型基于L1/2正则化理论建立,有效改善传统模型存在的变量选择与计算过拟合问题.文中算法基于"坐标下降"思想构造,快速有效.在一系列人工和实际数据集上的实验表明,文中算法在分类问题中具有良好的变量选择能力和预测能力,优于传统Logistic回归和L1正则化Logistic回归.     

基于机器学习的肿瘤免疫治疗应答预测研究

肿瘤免疫治疗应答的预测对肿瘤治疗方案设计及治疗有着重要的意义。本文引入基于随机森林的机器学习方法,将病人黑色素瘤组织转录组RNA-seq的基因表达谱作为特征,对免疫检查点阻断治疗的结果进行预测研究。对病人的基因表达谱使用随机森林算法来构建预测模型,并与Logistic回归模型和XGBoost模型进行比较。实验结果表明,随机森林模型对免疫检查点阻断治疗的应答能够进行较准确的预测,并且较Logistic回归模型和XGBoost模型预测效果更好。     

The relationship between perceived e-service quality and brand equity: A simultaneous equations system approach

Using the data collected from the online banking users in Taiwan, we build a hierarchical model of e-banking service quality and investigate the relationships among e-service quality, trust, satisfaction, loyalty, and brand equity. Moreover, the simultaneous equations system approach is also applied to transfer the traditional satisfaction-loyalty path into the simultaneous relationships between trust and loyalty, and between satisfaction and loyalty. As the structural form coefficients demonstrate the direct relations between our research constructs, the reduced form estimates further disclose the total impacts of the quality of e-banking service on trust, satisfaction, loyalty, and brand equity, respectively. The results indicate that the perceived quality formed through interaction with an online banking service positively affects customer trust and satisfaction, which in turn influence loyalty and brand equity. The significant simultaneous relationships between trust and loyalty, and between satisfaction and loyalty, are confirmed by our data, implying that these relationships must be determined simultaneously, rather than sequentially.     

How to Improve E‐Satisfaction and E‐Loyalty and Strengthen the Links Between Them: Value From Regulatory Fit

This article aims to investigate how regulatory fit can improve e‐satisfaction and e‐loyalty and strengthen the links between e‐satisfaction and both its antecedents (two technology acceptance model factors and the perceived quality of e‐shopping) and consequence (e‐loyalty). The research model and hypotheses are constructed through a literature review. An empirical study is performed to test the proposed research model, using survey research. The data are gathered via a questionnaire, which is developed on the basis of prior empirical studies. Results from this study point to the following: first, the two technology acceptance model factors and the perceived quality of e‐shopping significantly affect e‐satisfaction, which in turn e‐loyalty. Second, regulatory fit not only improves e‐satisfaction and e‐loyalty but also strengthens the links between e‐satisfaction and both its antecedents and consequence. On the basis of these findings, the implications are discussed and directions for future research are highlighted.     

Understanding customers' loyalty intentions towards online shopping: an integration of technology acceptance model and fairness theory

As with any other information system (IS), the success of online shopping depends largely on customer satisfaction and other factors that will eventually increase customers' loyalty intentions. This article integrates two major variables of technology acceptance model (TAM), trust, and fairness to construct a model for investigating the motivations behind customers' loyalty intentions towards online shopping. The hypothesised model is validated empirically using data collected from 311 customers of an online shopping store. The results indicated that distributive, procedural and interactional fairness were strong predictors of trust, which in turn influenced satisfaction. Distributive fairness and interactional fairness exhibited significant positive effects on satisfaction. Perceived usefulness and satisfaction influenced loyalty intention towards online shopping. Perceived ease of use acts indirectly on loyalty intention through the mediating effect of perceived usefulness. Implications for theory and practice and future research directions are discussed.     

Consumer perception of interface quality, security, and loyalty in electronic commerce

Customer interface quality, perceived security, and customer loyalty are critical factors for success of an e-commerce website; however, the relationships among them are not fully understood. We proposed a model for testing the relationships among them and the important outcomes of the site: switching costs and customer loyalty. Data was collected to test the model using a web-based survey, and empirical analyses were performed using SEM. The analytical results demonstrated that customer interface quality and perceived security positively affected customer satisfaction and switching costs, and thus customer loyalty to an e-commerce website. Specifically, our study showed that switching costs positively moderated the effect of customer satisfaction on customer loyalty; this moderating effect is discussed.     

Why do not satisfied consumers show reuse behavior? The context of online games

Empirical studies have consistently identified consumer satisfaction as a key antecedent to reuse behavior. Business has focused on consumer satisfaction as a way to improve consumer loyalty. However, it remains unknown why there are so many satisfied consumers who do not show reuse behavior. Therefore, the present study examined the potential factors moderating the association between consumer satisfaction and reuse behavior. A literature review and focus groups were used to explore moderating variables, and hierarchical regression analysis was used to test their influence. Results from a study of 490 online gamers in Taiwan showed that the relationship between consumer satisfaction and reuse behavior is contingent upon the moderating effects of customer involvement and perceived specific/upgrading activities. In particular, the relationship between consumer satisfaction and reuse behavior is more sensitive when customer involvement is lower.     

Are customer satisfaction and customer loyalty drivers of customer lifetime value in mobile data services: a comparative cross-country study

Many research papers have been published on the effect of customer satisfaction and customer loyalty on customer profitability which is related to customer lifetime value (CLV). However, there is limited research on the impact of cross-cultural factors on the effect of customer satisfaction and customer loyalty on CLV. This study aims to fill this gap. Focusing on the usage of mobile data services, 846 samples from China and 689 from the US are obtained. Data analysis suggests that customer loyalty is a driver of CLV, while customer satisfaction is not. This research has important implications for firms about how to enhance CLV in mobile data services.     

The Mechanisms through Which Certain Variables Influence Customer Loyalty: The Mediating Roles of Perceived Value and Satisfaction

This study surveyed 123 bank users who experienced problems with their banks’ online banking services, with the aim of understanding the mechanisms through which certain variables influence customer loyalty. Principally, the possible mediating roles of value and satisfaction on the relationships between recovery and loyalty and between ISO 9001 and loyalty were scrutinized. The dimensionalities of the scales were assessed using exploratory and confirmatory factor analysis. Thereafter, structural equation modeling and multiregression analyses were used to test the proposed model. The overall results showed that service recovery can be used to enhance customers’ satisfaction and perception of value and, therefore, customer loyalty. This study also confirmed the partial mediating roles of customers’ satisfaction and perception of value in the relationship between service recovery and loyalty. In contrast, ISO 9001 has no influence on perceived value, satisfaction, and loyalty; the mediating roles of both perceived value and satisfaction were not supported in the relationship between ISO 9001 and loyalty. In practice, even though ISO 9001 appeared to offer few extras to satisfy and retain customers, its use is still advisable because of the other potential benefits that it provides.     

A stochastic model on the profitability of loyalty programs

Effectiveness of customers’ loyalty programs has been the focal point of some recent studies. While empirical research shows mixed findings, analytical studies on the efficacy of loyalty programs are in their early stages. In this paper, we develop an analytical model on the profitability of loyalty programs in which customers’ valuations along with their satisfaction levels are incorporated as stochastic variables. The model consists of a revenue-maximizing firm selling a product through two periods. A loyalty reward is offered to two-period buyers in the form of an absolute-value discount on the price in the second period. The satisfaction level is represented by the difference between a customer’s original and post-purchase valuation. The formulation yields a stochastic programming problem with a nonlinear non-convex objective function. The problem is solved in terms of the model parameters. The results reveal that depending on the mean and variance of the satisfaction levels, the firm may be better off not to offer a loyalty reward. Specifically, if the mean of satisfaction levels turns out to be positive with a coefficient of variation less than a certain threshold, not adopting the loyalty program is optimal.     

The antecedents of consumers’ loyalty toward Internet Service Providers

Internet popularity is growing at an impressive rate. Sooner or later, every consumer comes face to face with the decision of choosing an Internet Service Provider (ISP). This study developed and empirically tested a model examining the antecedents of consumer loyalty toward ISPs. It incorporated the concept of expected technology change into the model to examine its effects on the formation of ISP loyalty intention. Confirmatory Factor Analysis (CFA) was performed to examine the reliability and validity of the measurement model, and the structural equation modeling techniques were used to evaluate the casual model. Based on an internet survey in Taiwan, this study showed that perceived value is very important in generating overall customer satisfaction and loyalty intention toward an ISP, and that perceived trust of an ISP enhances perceived value, overall satisfaction, and loyalty intention. However, the study demonstrated that future ISP technology expectancy exerted a negative influence on a consumer’s overall satisfaction and loyalty intention toward their ISP.     

The role of perceived usability,reputation, satisfaction and consumer familiarity on the website loyalty formation process

The increasing competitiveness in electronic business is motivating varying research on the principal antecedents of consumer loyalty. However, previous research is too scarce, so that there are notable aspects which must be analyzed in-depth. This research has focused on new topics about the development of online consumer loyalty, such as the role of familiarity and usability. Firstly, we analyze the effect of usability, reputation and satisfaction on consumer loyalty. The results have revealed the existence of a positive and significant influence of reputation and satisfaction on website loyalty. Secondly, we confirm the influence of usability on consumer satisfaction. Thirdly, we assess the moderating role of consumer familiarity on the relationships proposed in the research model. The analyses have shown that usability plays a special role in the loyalty formation process. At reduced levels of familiarity usability influences on loyalty through consumer satisfaction (indirect effect). However, if familiarity increases then perceived usability influences on loyalty not only through its effect on satisfaction, but also in a direct way. Consequently, higher levels of familiarity mean that the weight of usability in the explanation of website loyalty is higher. Finally, several conclusions, managerial suggestions, limitations and future research lines are proposed.