Exploiting multi-context analysis in semantic image classification

INTRODUCTION The popularity of digital images is rapidly in- creasing due to improving digital imaging technolo- gies, and convenient availability facilitated by the Internet. Organizing these images into categories and providing effective indexing is imperative for real- time browsing and retrieval. Typically, existing im- age classification work such as that of Vailaya et al.(2001) follows the paradigm of content-based im- age retrieval (CBIR) technologies, i.e., representing images u…     

面向信息处理的“二标三句式”复句层次关系判定

复句层次关系识别对计算语言学句法分析和复句信息工程意义重大。为解决关系标记显现不充足给复句层次关系识别所带来的困扰,需要充分挖掘可供计算机理解的复句关联的形式化句法语义知识。本文探讨了分句主语指称一致性和谓语语义相关性对分句语义关联的影响,深入分析了4种联结模式的"二标三句式"复句的句法语义特点,总结提取出11条层次关系判定规则,以期为复句层次关系识别奠定基础。语例验证表明,该研究能有效解决"二标三句式"复句的层次关系判定问题。     

Benefits left on the table: Evidence from the Servicemembers’ Civil Relief Act

In this study we test whether behaviorally-motivated informational interventions can lead individuals to utilize consumer financial protections. Specifically, we use a large-scale randomized controlled trial to test whether gain/loss framing and reminder messaging can lead servicemembers in the United States Army to utilize an interest-rate protection outlined in the Servicemember Civil Relief Act (SCRA). While we find that reminder messaging increases engagement with informational materials, we find no differences in engagement across gain and loss framing. Furthermore, we find that information provision has no effect on servicemember credit outcomes, regardless of how SCRA protections are framed or whether email reminders are sent. We run a second experiment to explore what factors may be limiting the efficacy of our interventions and find that low engagement with emails and significant attrition throughout the application process are likely to contribute to our results. Taken together, our results suggest that financial education or information interventions offered via email face significant challenges to their effectiveness even when the apparent benefits seem large.     

National trends in changes in special education classification among pre-elementary education children

Classification changes are common in special education. Using the first four years of the Pre-elementary Education Longitudinal Study data set (N = 3000), we investigated national trends in classification changes among young children with disabilities, the relationship between classification changes and children’s demographic information, and the relations between classification changes and children’s performance outcomes over time. The results revealed that declassification rates ranged from 35.6% for the three-year-old cohort to 46.3% for the five-year-old cohort. Approximately 25% to 35% of the children in each age cohort were reclassified at least once. The results showed that the prevalence rates of de/reclassification differed according to children’s demographic characteristics without consistent pattern. The study also revealed that reclassified children consistently exhibited lower performance than those without classification changes over time. The results suggest that reclassification did not help children “catch up” with those who remained in the same disability categories.     

基于网络语义标注的图像内容分析与理解

随着网络的迅速发展和应用,网络语义标签已经开始广泛地用于图像内容的标注和分享。由于图像本身及不同主体对图像的不同理解会造成图像分析理解的差异,因此如何充分利用网络标签对图像进行准确分析理解成为本文主要研究内容。图像特征包含了图像本身的大量内容信息,为了能够建立图像内容信息与网络标签之间的关系,本文主要工作包括：1,建立低层特征与图像间相似性关系;2,建立基于随机漫步模型平衡图像内容及网络标签间的关系以达到准确对图像内容分析和理解效果。实验结果表明本文所提出方法的有效性和可行性。     

Deep web站点查询界面的潜在语义分析

为了进一步提高搜索引擎的效率,实现对deep web中所蕴含的大量有用信息的检索、索引和定位,引入潜在语义分析理论是一种简单而有效的方法.通过对作为deep web站点入口的查询界面里的表单属性进行潜在语义分析,从表单属性中挖掘出潜在语义结构,并实现一定程度上的降维.利用这种潜在语义结构,推断对应站点的数据内容并改善不同站点的相似度计算.实验结果显示,潜在语义分析修正和改善了deep web站点的表单属性的语义理解,弥补了单纯的关键字匹配带来的一些不足.该方法可以被用来实现为某一站点查找网络上相似度高的站点及通过键入表单属性给出拥有相似表单的站点列表.     

基于 LSTM-Attention 的农业短文本信息分类研究

农业短文本中包含词数较少,导致语义获取不充分和分类效果下降。利用 Attention 机制加强关键词在分类时的权重,并结合 BiLSTM 设计 LSTM-Attention 模型。对 30 000 份原始数据经过中文分词、句法分析、文本向量化后,将 LSTM-Attention 模型训练成一个 LSTM-Attention 分类器,解决分类器对待分类文本数据敏感的问题。利用 30 000 份标准数据和加 30%干扰信息的复杂数据测试分类器分类效果,结果表明,LSTM-Attention 模型分类正确率达 98.59%,比传统 LSTM 模型高 3.72%,比 BiLSTM 模型高 1.61%,说明使用 BiLSTM 结 合 Attention 机制能够有效提升农业短文本分类效果。利用不同测试数据对 LSTM-Attention 分类器测试发现,LSTM-Attention 分类器具有良好收敛性,其分类效果不依赖于分类数据特征,分类效果稳定性佳。     

基于分类本体的web集成

在web信息集成领域,为消除语义异构、实现语义融合,将分类本体引入WWW信息集成,设计了一种基于本体集成的web信息集成系统.通过构建标准分类本体以获取局部分类本体,并建立二者间的映射,以获得多源统一视图.通过计算概念间的统领匹配度、关联匹配度、从属匹配度来完成概念的映射.实现了基于分类本体的web信息集成系统,该系统能很好地解决web信息分类语义异构问题,并能实现多web信息源的集成以及用户个性化定制.     

基于 DenseNet 的人脸脸型分类研究

脸型在三维人脸重建、人脸识别与检索等领域有着重要应用。针对脸型分类问题提出一种基于傅里叶描述符、三角形半径描述符和几何特征描述符的 DenseNet 网络脸型分类方法。通过主动形状模型方法定位得到人脸边缘轮廓点,分别使用傅里叶描述符、三角形半径描述符和几何特征描述符将轮廓点数值化为一维向量,并将其输入 DenseNet 网络进行训练,实现脸型分类。通过实验对比分析,该方法能够充分挖掘人脸形状信息,有效提高脸型分类准确率,同时避免旋转、尺度等影响。     

甲骨文领域语义词典的构建研究

随着甲骨文数字化研究的不断深入,建立综合的甲骨文语义知识库已成为当前甲骨文研究的迫切要求,而语义词典则是构建语义知识库的基础。本文从甲骨文语言自身特点出发,首先明确甲骨文语义分析原则,给出了甲骨文语义分类体系;然后按照配价原则描述对应的语义属性及语义搭配;最后构造一个半自动化的甲骨文语义词典生成系统,以期更好地服务于甲骨学研究。     

基于XML/RDF和Ontology的语义网之构建

语义网能够提高计算机对各种信息的分析和处理能力，是下一代互联网的一个发展方向。介绍了语义网的基本概念及作用，并提出综合运用XML/RDF体系和本体论来构建语义网。     

基于微博文本的用户人格分析模型研究

传统的微博用户人格分析将人格分为五类,但未考虑人格类别之间潜在的关联性。为此基于多标签集成分类方法（RAkEL）进行改进,构建RAkEL-PA模型。RAkEL-PA模型使用标签集合中不同的随机子集训练相应的Label Powerset（LP）分类器,然后集成所有分类结果作为最终分类结果。在微博用户文本消息数据上进行实验,结果表明,RAkEL-PA模型的两个不同策略对用户人格分类准确率较高。RAkEL-PA模型充分考虑多个人格之间的相关性,以提高用户人格分类鲁棒性。     

A Predictive Model of Inquiry to Enrollment

The purpose of this paper is to build a predictive model of enrollment that provides data driven analysis to improve undergraduate recruitment efforts. We utilize an inquiry model, which examines the enrollment decisions of students that have made contact with our institution, a medium sized, public, Doctoral I university. A student, who makes an inquiry to our university such as by returning a request for information form, often provides far less information than is available from applicants. Despite this fact we find that characteristics of the student, as well as geographic and demographic data based on the student’s zip code are significant predictors of enrollment. Accounting for uncertainty in our model’s specification, we find that we are able to predict out of sample the enrollment decision of 89% of student inquiries. We also demonstrate how these findings can be used to improve marketing efforts.     

句子理解中语言信息的整合加工机制

句子理解过程是对语音、词汇、句法、语义层面语言信息的整合加工过程。分段式、并行式和一体式模型争论的核心在于语言信息加工的时间进程和交互作用。根据已有实验推测：1)语言信息在句子加工中起作用的方向很　可能由其本身层次属性决定：句法信息作用方向可能向后,语义信息作用方向可能向前;2)语言信息在句子加工中起作用的强度也很有可能由其本身层次属性决定,但都还需进一步实验验证。语用层面信息加工过程的研究需要加强,实验材料为汉语的相关研究有待单独梳理。     

一种用于文本分类的特征项权值计算方法的研究

文章考虑特征项的概率信息又结合文本的语义信息来计算特征项的权值,提出一种新的用于文本分类的特征项权值计算方法（FreSem）,采用支持向量机（SVM）分类器进行实验,与传统的频率、TFIDF两种方法相比,能有效地提高文本分类的正确率。     

A canonical correlation analysis of demographic information and the behavior problem checklist

The purpose of the present study was to do a canonical correlation analysis of demographic information and the Behavior Problem Checklist (BPCL) in order (a) to employ the demographic information in a manner that had not been possible using other methods, and (b) to examine the four factors on the BPCL as a general measure of mental health and mental illness. Subjects were 2,314 children who differed in grade in school, sex, social class, ethnic group status, and type of class in school (i.e., regular or special education). Teachers provided the demographic information and the BPCL ratings. It was found that the four factors on the BPCL were significantly correlated. It was also found that there were four canonical correlations between the demographic information and the BPCL ratings of which three were and one was not significant. The meanings of the four canonical correlations were described for the demographic information and for the four factors on the BPCL.     

高校科技英语类文章的阅读技巧

科技英语类文章在英语阅读材料中居于偏难的程度,其句式严谨,语言具有逻辑性,同时词汇专业性强,不过,再长再难的句子,仔细推敲,还是有一定的阅读技巧,该文按照信息获取,分类,整合的步骤,介绍了一种行之相对有效的方法,可以缓解读者对于此类文章的困扰。     

Developmental Change in Infant Categorization: The Perception of Correlations among Facial Features

Previous studies indicated that the ability to detect correlations among attributes emerges between 7 and 10 months of age. In the present study, the generality of this developmental transition was examined. Using an infant-control habituation procedure, 48 7- and 10-month-old infants were tested for the perception of correlations among basic facial features. The developmental effects were replicated. Only the 10-month-old infants demonstrated their sensitivity to the pattern of correlation by generalizing to a novel face that preserved the experienced pattern of correlation, while showing increased attention to a face in which the pattern of correlation was violated. 7-month-old infants generalized to both test stimuli containing familiar features, regardless of the status of the correlation. Implications for face perception and the processing of categorical information are discussed.     

FARMER: A novel approach to file access correlation mining and evaluation reference model

File semantic has proven effective in optimizing large scale distributed file system. As a consequence of the elaborate and rich I/O interfaces between upper layer applications and file systems, file system can provide useful and insightful information about semantic. Hence, file semantic mining has become an increasingly important practice in both engineering and research community. Unfortunately, it is a challenge to exploit file semantic knowledge because a variety of factors could affect this information exploration process. Even worse, the challenges are exacerbated due to the intricate interdependency between these factors, and make it difficult to fully exploit the potentially important correlation among various semantic knowledges. This article proposes a file access correlation miming and evaluation reference (FARMER) model, where file is treated as a multivariate vector space, and each item within the vector corresponds a separate factor of the given file. The selection of factor depends on the application, examples of factors are file path, creator and executing program. If one particular factor occurs in both files, its value is non-zero. It is clear that the extent of inter-file relationships can be measured based on the likeness of their factor values in the semantic vectors. Benefit from this model, FARMER represents files as structured vectors of identifiers, and basic vector operations can be leveraged to quantify file correlation between two file vectors. FARMER model leverages linear regression model to estimate the strength of the relationship between file correlation and a set of influencing factors so that the “bad knowledge” can be filtered out. To demonstrate the ability of new FARMER model, FARMER is incorporated into a real large-scale object-based storage system as a case study to dynamically infer file correlations. In addition FARMER-enabled optimize service for metadata prefetching algorithm and object data layout algorithm is implemented. Experimental results show that is FARMER-enabled prefetching algorithm is shown to reduce the metadata operations latency by approximately 30%–40% when compared to a state-of-the-art metadata prefetching algorithm and a commonly used replacement policy.     

集情绪与性别二维信息于一体的实验材料标准化研究

目的:收集整理一套集情绪信息与性别信息于一体的标准化实验材料,从而更精细地研究情绪信息、性别信息(语义信息)加工及二者的交互作用。方法:通过方便取样选取40名研究生,对从网络收集和由Facegen modller 3.5软件制作并统一处理后的401张情绪和性别信息均明确以及二者信息均模糊的面孔图片,在男、女、积极、消极、吸引注意的程度、唤醒程度六个信息特征上用七点量表进行评分;对97个具有情绪意义或性别意义的共四类单字分别在积极、消极、男、女意义特征上用七点量表进行评分。结果:该材料库中最后有255张面孔图片的情绪信息和性别信息的强强特征无差异(p0.05),弱弱特征无差异(p0.05),强弱特征差异显著(p0.05);在笔画数和字频平衡后,男、女、积极、消极意义共四类92个字的强特征值和弱特征值与4进行的单样本t检验结果为,强特征t(91)=83.127,p0.001,弱特征t(91)=-96.124,p0.001。结论:严格量化匹配了情绪信息和性别(语义)信息的面孔图片,充实了可用于表情加工、性别认知、面孔识别等研究领域的图片库,其研制思路对于情绪与认知的关系研究亦会有所启示。