融合局部低秩先验与Bloch流形约束的磁共振指纹重建算法

为了实现快速成像，磁共振指纹（Magnetic Resonance Fingerprinting,MRF）技术通常使用非笛卡尔稀疏采样模板对K空间进行高度欠采样，从而获得稀疏K空间信号.然而，从稀疏的K空间信号重建像空间数据是一个病态不适定问题，重建出的MRF像空间数据存在大量的混叠伪影，直接影响到组织生理参数的重建准确度.为此需要将各种先验知识引入重建模型之中，以缓解MRF重建问题的不适定性.针对上述问题，本文提出一种融合局部低秩先验与Bloch流形约束的MRF重建模型，并使用交替方向乘子法（Alternating Direction Method of Multipliers,ADMM）求解模型中的非凸MRF重建问题.本文算法在引入MRF像空间数据的局部低秩先验的同时，使用预先构建的字典为重建指纹提供流形约束.一方面通过空域局部低秩约束有效抑制混叠伪影的产生，另一方面利用字典先验避免指纹的时域流形特征在迭代重建过程中丢失.仿真实验结果表明，相较于引入了全局低秩先验与Bloch流形约束的其他同类算法，本文算法可以提供更高的组织参数重建准确度.     

基于多维云概念嵌入的变分图自编码器研究

变分图自编码器是图嵌入研究中重要的深度学习模型,但存在着先验正态分布缺陷、训练过程中容易出现后验塌陷等问题．本文从建立云概念空间与隐空间的映射关系入手,引入云模型数字特征对网络中的节点进行不确定性概念表示,设计了一种基于多维云模型的变分图自编码器（Variational Graph Autoencoder based on Multidimensional Cloud Model,MCM-VGAE）．该模型实现了隐空间的多维云概念嵌入及相应的漂移性损失度量,将先验分布扩展为泛正态分布,利用多维正向云发生器及云包络带修正采样算法实现了重参数化过程,有效缓解了后验塌陷现象．在应用效果上,模型在多类型数据集上的链路预测、节点聚类、图嵌入可视化实验表现均优于基准模型,进一步说明了方法的普适有效性．     

基于稀疏贝叶斯学习的大规模多用户检测算法

针对现有算法大都基于高斯逆伽马先验模型的稀疏贝叶斯学习（GIG-SBL），忽略了稀疏解所对应的支撑集向量稀疏性的问题，提出一种基于伯努利高斯逆伽马先验模型的稀疏贝叶斯学习（BGIG-SBL）架构，通过引入一个伯努利先验的二元向量，设计了单测量向量（SMV）的BGIG-SBL-SMV算法，结合支撑集向量的稀疏性提高重构性能。进一步将该算法扩展到多测量向量（MMV）方案，通过共享相同控制稀疏解的超参数，利用MMV的联合稀疏性提出BGIG-SBL-MMV算法。实验结果表明，所提BGIG-SBL-SMV算法相较于传统GIG-SBL-SMV算法，在mMTC用户检测场景可实现2 dB的性能增益；同时，所提BGIG-SBL-MMV算法相对于单测量向量BGIG-SBL-SMV算法，用户检测率和数据检错率的性能增益可达到4 dB，证明了所提算法的优越性。     

基于短语向量和主题加权的关键词抽取方法

现有关键词抽取算法缺乏对短语的有效表示,为抽取出更能反映文本主题的关键短语,本文提出一种基于短语向量的关键词抽取方法PhraseVecRank.首先设计基于LSTM(Long Short-Term Memory)和CNN(Convolutional Neural Network)自编码器的短语向量构建模型,解决复杂短语的语义表示问题.然后,利用短语向量对每个候选短语计算主题权重,通过主题加权排序提高关键词抽取的效果.在公共数据集和学术论文数据上的实验表明,本文提出的方法能够有效提取与文本主题信息相关的关键短语,同时利用自编码器构造的短语向量可以更好地表示短语的语义信息.     

社区环境下基于节点交互和主题的影响力计算模型

为解决现有算法对社交网络节点影响力计算准确度不高的问题，本文整合节点不同维度信息，综合考虑节点在多个主题社区上的主题分布向量，提出一种新的节点影响力计算模型.模型首先将主题相关性作为先验信息；然后利用混合隶属度随机块（Mixed Membership Stochastic Block）模型表达节点间的交互关系，用主题模型学习主题内容；最后结合全局拓扑关系迭代计算节点的全局影响力.本文选取社交网络数据，以P@N、MAP等作为评价指标同现有主流算法进行比较.实验结果显示，本文算法有效提升了影响力节点识别的准确度和排名的有效性.     

基于专家场先验模型的图像超分辨重建算法

为了进一步提高图像超分辨率重建的质量,针对非局部集中稀疏表示算法中重建图像的噪声问题,提出了一种基于专家场先验模型的图像超分辨率重建改进算法。首先,利用专家场模型从图像训练集中学习整幅图像的先验知识建立全局先验模型;然后将学习到的先验信息用于非局部集中稀疏表示模型求解最优稀疏表示系数;最后,得到高分辨率图像估计。该算法在超分辨率重建迭代运算的同时,同步更新专家场模型参数,因此在不显著增加运算复杂度的情况下,通过选取合适的先验约束,有效地增强了图像重建的效果。实验结果表明:相比非局部集中稀疏表示算法,文中算法对无噪和有噪降质图像均能取得较好的峰值信噪比结果,并且能够进一步提高有噪图像的去噪效果。     

基于SVM算法的在线评论情感分类研究

在线评论情感分析是商户和消费者共同关注的热点,基于词典的传统情感分类方法不适用于在线评论的分类,因此提出基于SVM算法的在线评论情感分类模型.首先通过清洗、分词、标注情感标签对在线评论进行预处理,然后进行词向量表示,最后使用SVM算法进行分类.实验结果表明,该模型具有较为理想的分类准确率.     

基于用户模型的加权位置/摘要排序算法

为了提高查询精度,提出了一种个性化元搜索引擎模型．在该模型中,引入了一种基于用户兴趣模型的加权位置／摘要的查询结果排序算法,该算法综合考虑了查询结果的摘要与查询的全局相关度、查询结果在各个成员搜索引擎返回结果中的排列位置信息、各个成员搜索引擎对查询意图主题类别的相关度三方面的信息．实验表明该排序算法能较好地把与用户查询意图相关的结果排在查询结果的最前面,提高了查准率．     

基于知识图谱表示学习的谣言早期检测方法

社交网络谣言是严重危害社会安全的一个重要问题.目前的谣言检测方法基本上都依赖用户评论数据.为了获取可供模型训练的足量评论数据，需要任由谣言在社交平台上传播一段时间，这就扩大了谣言的危害.本文提出了一种基于知识图谱表示学习的谣言检测方法.该方法不依赖用户评论数据.首先基于PN-KG2REC算法得到实体和关系的表示；然后将待检测三元组中的实体和关系表示进行拼接，得到三元组表示；最后对三元组的向量表示进行分类，并根据分类结果判断待检测三元组描述内容的真假性.采用公开数据的实验结果表明，本文提出的谣言检测方法在不依赖用户评论数据的前提下，能够有效地对谣言进行早期检测.     

一种基于模型的可分解贝叶斯在线强化学习

针对贝叶斯强化学习中参数个数巨大,收敛速度慢,无法实现在线学习的问题,提出一种基于模型的可分解贝叶斯强化学习方法.首先,将学习参数进行可分解表示,降低学习参数的个数;然后,根据先验知识和观察数据采用贝叶斯方法来学习,最优化探索和利用二者之间的平衡关系;最后,采用基于点的贝叶斯强化学习方法实现学习过程的快速收敛,从而达到在线学习的目的.仿真结果表明该算法能够满足实时系统性能的要求.     

How online ratings and trust influence health consumers’ physician selection intentions: An experimental study

Consumers’ growing reliance on the web information when choosing physicians has drawn attention to the need for more research into online physician ratings. This study conducted an experimental design to explore the roles of trust in different stages and three key online rating characteristics, including overall numerical ratings (high or general), review volume (high or low), and a comparison of the effect rating and attitude rating (higher or lower effect rating than attitude rating), in influencing health consumers’ choices. Results suggested that the overall numerical rating and review volume were significantly and positively correlated with the physician selection intention. Perceived physician trustworthiness completely mediated the effect of the review volume on consumer intentions, while initial trust in online physician ratings produced a moderating effect. The comparison of sub-dimension rating scores induces consumers’ regulatory focus and further moderated the relationship between overall rating score and consumers’ selection intentions, as well as the relationship between perceived physician trustworthiness and consumers’ selection intentions. Implications, limitations and future research directions are also discussed.     

基于主题模型的博客标签语义知识获取

This paper focuses on semantic knowledge acquisition from blogs with the proposed tag-topic model. The model extends the Latent Dirichlet Allocation (LDA) model by adding a tag layer between the document and the topic. Each document is represented by a mixture of tags; each tag is associated with a multinomial distribution over topics and each topic is associated with a multinomial distribution over words. After parameter estimation, the tags are used to describe the underlying topics. Thus the latent semantic knowledge within the topics could be represented explicitly. The tags are treated as concepts, and the top-N words from the top topics are selected as related words of the concepts. Then PMI-IR is employed to compute the relatedness between each tag-word pair and noisy words with low correlation removed to improve the quality of the semantic knowledge. Experiment results show that the proposed method can effectively capture semantic knowledge, especially the polyseme and synonym.     

面向顾客点评数据的属性层次观点挖掘研究

This paper focuses on how to improve aspect-level opinion mining for online customer reviews. We first propose a novel generative topic model, the Joint Aspect/Sen-timent (JAS) model, to jointly extract aspects and aspect-dependent sentiment lexicons from online customer reviews. An as-pect-dependent sentiment lexicon refers to the aspect-specific opinion words along with their aspect-aware sentiment polarities with respect to a specific aspect. We then apply the extracted aspect- dependent sentiment lexi-cons to a series of aspect-level opinion mining tasks, including implicit aspect identification, aspect-based extractive opinion summarization, and aspect-level sentiment classification. Experimental results demonstrate the effectiveness of the JAS model in learning aspect- dependent sentiment lexicons and the practical values of the extracted lexicons when applied to these practical tasks.     

Assessing the helpfulness of online hotel reviews: A classification-based approach

With the rapid development of Web 2.0, travelers have started sharing their travel experiences on websites. The expanding amount of online hotel reviews results in the problem of information overload. Therefore, the effective identification of helpful reviews has become an important research issue. In this study, online hotel reviews were collected from TripAdvisor.com, and the helpfulness of these reviews was comprehensively investigated from the aspects of review quality, review sentiment, and reviewer characteristics. Review helpfulness prediction models were also developed by using classification techniques. The results indicate that reviewer characteristics are good predictors of review helpfulness, whereas review quality and review sentiment are poor predictors of review helpfulness.     

Can book covers help predict bestsellers using machine learning approaches?

As the book publishing market changes from offline to online, readers tend to purchase books while paying more attention to book covers and metadata rather than the actual book contents. We examine whether publishers can know users’ satisfaction with books in advance, and both metadata and book covers help predict this satisfaction. Exploring effects of metadata and book covers on the satisfaction is not only necessary for publishers’ perspectives, but also for librarians’ perceptions. However, the majority of prior research on user preference-based book recommendation systems in both book industry and library system employed review comments, ratings, or book loan records. Thus, we open up the potentiality of other factors, which implicitly affect the satisfaction with books. We collected book titles, authors, publishers, reviews, ratings, and covers from the “Literature and Fiction” genre in the Amazon bookstore and conducted an experiment to predict readers’ satisfaction ratings based on book reviews, metadata, and book covers. Several deep learning classifiers (CNN, ResNet, LSTM, BiLSTM, GRU, BiGRU) were employed. Reviews alone can reach a certain level of prediction performance, but adding metadata, cover images, and cover objects to a review-based predictive model slightly improves that performance. Based on these results, we confirmed that both metadata and book covers improve predicting readers’ perceived satisfaction. This study is a pilot exploration of the idea that multimodal approaches can improve the prediction of the perceived satisfaction of book readers. Moreover, we have publicly released both source codes and data samples employed in this study.     

What is the impact of service quality on customers’ satisfaction during COVID-19 outbreak? New findings from online reviews analysis

The COVID-19 pandemic has caused major global changes both in the areas of healthcare and economics. This pandemic has led, mainly due to conditions related to confinement, to major changes in consumer habits and behaviors. Although there have been several studies on the analysis of customers’ satisfaction through survey-based and online customers’ reviews, the impact of COVID-19 on customers' satisfaction has not been investigated so far. It is important to investigate dimensions of satisfaction from the online customers’ reviews to reveal their preferences on the hotels' services during the COVID-19 outbreak. This study aims to reveal the travelers’ satisfaction in Malaysian hotels during the COVID-19 outbreak through online customers’ reviews. In addition, this study investigates whether service quality during COVID-19 has an impact on hotel performance criteria and consequently customers' satisfaction. Accordingly, we develop a new method through machine learning approaches. The method is developed using text mining, clustering, and prediction learning techniques. We use Latent Dirichlet Allocation (LDA) for big data analysis to identify the voice-of-the-customer, Expectation-Maximization (EM) for clustering, and ANFIS for satisfaction level prediction. In addition, we use Higher-Order Singular Value Decomposition (HOSVD) for missing value imputation. The data was collected from TripAdvisor regarding the travelers’ concerns in the form of online reviews on the COVID-19 outbreak and numerical ratings on hotel services from different perspectives. The results from the analysis of online customers’ reviews revealed that service quality during COVID-19 has an impact on hotel performance criteria and consequently customers' satisfaction. In addition, the results showed that although the customers are always seeking hotels with better performance, they are also concerned with the quality of related services in the COVID-19 outbreak.     

An overview of online charging in 3GPP networks: new ways of utilizing user,network, and service‐related information

Modern charging systems routinely apply the user, network, and service‐related information while performing online charging. Compared, however, to all the information available to and used in managing the network as a whole, the charging systems only use a limited subset. This work is motivated by the challenge to identify which information is used, and how it is used in online charging‐related processes, and also to explore whether it could be utilized ‘better’ or ‘smarter’ to improve future online charging systems functionality. We do not attempt to predict which information will be utilized in such systems and for what purpose, but instead summarize the open issues in view of the emerging trend of exploiting the user, network and service‐related information in service provisioning. We focus on the most recent 3GPP standards and relevant research papers, and propose three key aspects of online charging, with respect to information utilization: (a) signaling aspect, (b) inter‐domain aspect, and (c) service‐ and component‐based aspect. We present a state‐of‐the‐art review by grouping the works found in the literature based on the aspects they are associated with, and compare them based on the proposed comparison criteria. The discussion presented at the end of the paper indicates three common open issues, namely: (1) lack of common charging information specification and structure; (2) lack of mechanisms for information sharing among stakeholders in the service delivery process; and (3) lack of a common framework for sharing information while protecting user privacy. Copyright © 2012 John Wiley & Sons, Ltd.     

Research on online review-trust model and synthetical value model based on power-law and density clustering algorithm

The existing online shopping researches on consumer reviews are mostly based on the attitude change model (ACM). Although the ACM is valuable, it is not easy to judge the trustworthiness of the reviews and measure the values of the reviews. Based on the online data acquisition technology, we have got the data of 360buy, a domestic large-scale business to customer (B2C) commerce website in China. With application of data-mining and the density clustering algorithm (DBSCAN), we focus on the intervals distribution and the synthetic value of consumer reviews. The distribution of review interval can be depicted by the power-law function which presents a monotonically increasing relationship between the power-exponent and the customers’ concerns with the commodity: the higher the exponent is, the more attention will be drawn. We also find that the value of online reviews can be measured by the expertise value, which is the attraction and the quality of the reviews. Based on the above results, we have constructed the online review-trust model and the synthetic value model. The relationship between the power-exponent and the consumer attention has played a vital role in the consumer attention to online-shopping, and then the synthetic value model will help people find out useful reviews more effectively.     

A data-driven approach to exploring similarities of tourist attractions through online reviews

The motivation for tourists to visit a city is often driven by the uniqueness of the attractions accessible within the region. The draw to these locations varies by visitor as some travellers are interested in a single specific attraction while others prefer thematic travel. Tourists today have access to detailed experiences of other visitors to these locations in the form of user-contributed text reviews, opinions, photographs, and videos, all contributed through online tourism platforms. The data available through these platforms offer a unique opportunity to examine the similarities and difference between these attractions, their cities, and the visitors that contribute the reviews. In this work, we take a data-driven approach to assessing similarity through textual analysis of user-contributed reviews, uncovering nuanced differences and similarities in the ways that reviewers write about attractions and cities.     

A Reputation Framework for Social Participatory Sensing Systems

Social participatory sensing is a newly proposed paradigm that tries to address the limitations of participatory sensing by leveraging online social networks as an infrastructure. A critical issue in the success of this paradigm is to assure the trustworthiness of contributions provided by participants. In this paper, we propose an application-agnostic reputation framework for social participatory sensing systems. Our framework considers both the quality of contribution and the trustworthiness level of participant within the social network. These two aspects are then combined via a fuzzy inference system to arrive at a final trust rating for a contribution. A reputation score is also calculated for each participant as a resultant of the trust ratings assigned to him. We adopt the utilization of PageRank algorithm as the building block for our reputation module. Extensive simulations demonstrate the efficacy of our framework in achieving high overall trust and assigning accurate reputation scores.