Doctor of Philosophy, The Ohio State University, 2012, Statistics

Linear Discriminant Analysis (LDA) has served as a standard technique in classification for many years. Many improvements have been proposed to boost the performance of LDA yet maintain its simplicity, intuitive appeal, and robust nature. Lotlikar and Kothari proposed fractional LDA (F-LDA) as an improved version of LDA. However, for a large number of classes, F-LDA is not feasible to implement in real time due to huge computational effort in the sequential search process for each dimension to be reduced. In addition, F-LDA is a directed projection pursuit technique, which takes several iterations to reduce just one dimension. Our research is focused on modifying these methods to be applicable to the face recognition problem (high-dimensional image data, large number of classes). Simultaneous Adaptive Fractional Discriminant Analysis (SAFDA) is a procedure developed specifically to learn a specified or fixed low-dimensional subspace in which classes are well separated by sequentially downweighting all dimensions to be removed simultaneously. Via analysis of a weighted between class scatter matrix of whitened data, the best projected space is learned through a directed projection pursuit method that focuses on class separation in the reduced space (rather than the full space like LDA). An adaptive kernel (Gaussian) was found to be the most suitable to avoid extra time considerations inherent in cross-validation by allowing the data to determine optimal bandwidth choice. While the SAFDA algorithm showed a marked improvement over standard LDA techniques in terms of classification, the additional computational time, compared to LDA, was minimal in situations involving a small number of classes (MNIST) as well as a large number of classes (AR face database). SAFDA also provides a procedure that matches (or in some cases, outperforms) the F-LDA benchmark in terms of classification, yet is much more feasible in computational time and effort.

Committee: Prem Goel PhD (Advisor); Radu Herbei PhD (Committee Member); Yoonkyung Lee PhD (Committee Member) Subjects: Statistics

Doctor of Philosophy, The Ohio State University, 2024, Business Administration

Text data is widely used in marketing research. This dissertation proposes two new methods that utilize text data in parallel and serial mediation analyses. A parallel mediation model that uses text data to identify multiple mediators in parallel mediation analysis is proposed. The model is based on the Latent Dirichlet Allocation (LDA) model that incorporates treatment and outcome variables. Treatment variables can affect topic composition in the text data, with topic probabilities used to predict outcomes via a logistic regression model. Lexical priors are introduced to seed topics that researchers consider relevant to an analysis, while non-seeded topics allow researchers to find other potential mediation paths. The resulting analysis of mediation replaces the use of rating scales with text that more flexibly reflects the reasons for respondent choices. The assessment of stimuli's effect on topic probabilities provides information on which aspects of stimuli contribute to the change in respondents' choices of words and their latent meanings behind these words. Consumers often engage in complex reasoning when exposed to new information contained in advertisements and websites. In this dissertation, a serial mediation method is proposed to understand consumers' thoughts about new information in a serial mediation framework using textual and fixed-point rating data. Treatment variables are assumed to affect the topic composition of the text data, which is then related to the rating data and an outcome variable. The proposed model flexibly identifies mediators and relationships in situations where scales are not well developed. Apart from the additional insights revealed from the textual data, the proposed model predictively outperforms existing models of mediation.

Committee: Greg Allenby (Committee Chair); Rebecca Walker Reczek (Committee Member); H. Alice Li (Committee Co-Chair) Subjects: Business Administration; Marketing; Statistics

Doctor of Philosophy (Ph.D.), Bowling Green State University, 2022, Photochemical Sciences

In the recent years, science has faced and is till facing numerous challenges associated with climate changes, environment pollution, food production, and access to inexpensive healthcare to name a few. These challenges have also created opportunities for fundamental research and development of new technologies that could alleviate the above problems. One example of such research activities is also the research and development of materials and methods useful for detecting and analyzing food or beverage components, environmental pollutants, or markers of important biological processes. Within this general field, molecules and materials that can be used as fluorescence-based sensors are very attractive. This is because of their potential for superior sensitivity at low cost and the fact that their operation does not require expensive equipment and highly trained personnel. Thus, the synthesis and investigation of supramolecular sensors that respond to analytes such as saccharides in food or beverage sources, or phosphates that are widely used as fertilizers and are also the source of eutrophication of water bodies are of both fundamental and practical importance. This dissertation describes the studies focused on supramolecular fluorescent chemosensors that are able to distinguish among various analytes that are structurally similar. We have synthesized and studied chemosensors for sensing of saccharides and their mixtures in beverages, as well as self-assembly based sensors for the detection of biologically important phosphate anions. Here, the sensing of phosphate-type anions is not only important for monitoring of eutrophication agents (orthophosphate and in lesser extent also oligo-/polyphosphates), but also for future application in biotechnologies such as quantitative polymerase chain reaction (qPCR). We hope that our contributions to research in sensors would contribute to further advancement of our knowledge and potential practical applications.

Committee: Pavel Anzenbacher Jr. Ph.D. (Committee Chair); Wendy Manning Ph.D (Other); Malcolm Forbes Ph.D (Committee Member); Jayaraman Sivaguru Ph.D (Committee Member) Subjects: Analytical Chemistry; Organic Chemistry

Doctor of Philosophy (Ph.D.), Bowling Green State University, 2022, Communication Studies

Digitalization becomes aggressively integrated into the policy agenda of modern nation-states arguably to accelerate their progress and impact democratization. Concurrently, digital surveillance is also growing worldwide. What happens to democracy when nation-states engage in such a paradoxical exercise of digitalization? This dissertation takes a fresh look at this problem in a transnational context and investigates the democratic implications of such digitalization practices. I examine the (un)changing development discourses within digital policy documents (N=41) and news articles (N=3,739) covering digitization in Bangladesh and Norway over 15 years (2003-2017). I specifically investigate the conceptual framing of three overarching elements of ICT4D — communication, technology, and development— using a new theoretical lens communication as critical freedom (CCF) that I propose uniting relevant works of Jurgen Habermas, Michell Foucault, and Amartya Sen. This inquiry explores how digital policy and news media discursively expand or limit democratization. An innovative mixed-method, computational-critical discourse analysis (C-CDA) is proposed and employed in doing the analysis, combining qualitative methods (i.e., critical discourse analysis) with computational techniques (i.e., LDA topic modeling). As the analyses suggest, Bangladesh and Norway advance a technocapital determinist logic of social change, which instrumentalizes “communication,” renders excessive agency to “technology,” and ultimately posits “development” as mere material progress. These nations' digital policy and news reports scrutinized in this study seem to have been shaped mainly by a transnational discourse of neoliberal globalization, making Bangladesh a digital proletariat and Norway a digital bourgeoisie in the spectrum of global development. Moreover, both nations are forging cybersecurity discourse as a new technique of power that legitimizes digital surveillance and control. Hence (open full item for complete abstract)

Committee: Srinivas Melkote Ph.D. (Advisor); Lara Lengel Ph.D. (Committee Member); Kei Nomaguchi Ph.D. (Other); Clayton Rosati Ph.D. (Committee Member); Syed Shahin Ph.D. (Committee Member) Subjects: Communication; Mass Communications; Mass Media

PhD, University of Cincinnati, 2018, Engineering and Applied Science: Computer Science and Engineering

The analysis of news content has been a central focus for media scholars, political scientists, sociologists and historians. Traditionally, it has been performed using relatively small news archives collected over limited periods. However, in the last 20 years, following the creation of the World-Wide Web, a dramatic change has occurred in the reporting and dissemination of news. One result of this change is that an ever growing archive of news is readily available as electronic text. This, in turn, is making it possible to analyze news on a large scale using methods developed in the fields of Web Intelligence, Data Mining and Machine Learning. The issues that news content analysis tries to address include: Identification of salient topics; summarization of stories; extraction of opinions; and characterization of news reports in terms of content, sentiment, bias, etc. These are also the motivating issues for this research. The research in this dissertation describes a framework called IONA: Intelligent Online News Analysis. This is meant to be a tool to accomplish four goals: 1) Extracting and visualizing important stories from real-time news streams; 2) Characterizing and comparing the cognitive/epistemic organization of all news in different media sources over the same time period; 3) Comparing the structure of specific stories from different media sources to characterize similarities, differences, and possible biases; and 4) Doing comparative analysis of how specific stories and the news streams from different media sources evolve over time in order to characterize the dynamics of news from each source. The IONA approach represents an innovative combination of methods from natural language processing, semantic analysis and complex networks. The identification of topics uses a novel algorithm that integrates Latent Dirichlet Allocation (\LDA) with tagging using Ngrams. The resulting topics are used to extract coherent sets of news reports from large corpora (open full item for complete abstract)

Committee: Ali Minai Ph.D. (Committee Chair); Raj Bhatnagar Ph.D. (Committee Member); Karen Davis Ph.D. (Committee Member); Carla Purdy Ph.D. (Committee Member); Anca Ralescu Ph.D. (Committee Member) Subjects: Computer Science

Doctor of Philosophy, The Ohio State University, 2018, Electrical and Computer Engineering

By combining different facial muscle actions, called Action Units (AUs), humans can produce an extraordinarily large number of facial expressions. Computational models and studies in cognitive science have long hypothesized the brain needs to visually interpret these action units to understand other people's actions and intentions. Surprisingly, no studies have identified the neural basis of the visual recognition of these action units. Here, using functional Magnetic Resonance Imaging (fMRI), we identify a consistent and differential coding of action units in the brain. Crucially, in a brain region thought to be responsible for the processing of changeable aspects of the face, pattern analysis could decode the presence of specific action units in an image. This coding was found to be consistent across people, facilitating the estimation of the perceived action units on participants not used to train the pattern analysis decoder. Research in face perception and emotion theory requires very large annotated databases of images of facial expressions of emotion. Useful annotations include AUs and their intensities, as well as emotion category. This process cannot be practically achieved manually. Herein, we present a novel computer vision algorithm to annotate a large database of a million images of facial expressions of emotion from the wild (i.e., face images downloaded from the Internet). We further use WordNet to download 1,000,000 images of facial expressions with associated emotion keywords from the Internet. The downloaded images are then automatically annotated with AUs, AU intensities and emotion categories by our algorithm. The result is a highly useful database that can be readily queried using semantic descriptions for applications in computer vision, affective computing, social and cognitive psychology. Color is a fundamental image feature of facial expressions. For example, when we furrow our eyebrows in anger, blood rushes in and a reddish color (open full item for complete abstract)

Committee: Aleix Martinez (Advisor); Julie Golomb (Committee Member); Yuan Zheng (Committee Member) Subjects: Cognitive Psychology; Computer Engineering; Computer Science; Social Psychology

Doctor of Philosophy, The Ohio State University, 2014, Statistics

There has been growing interest in kernel methods for classification, clustering and dimension reduction. For example, kernel linear discriminant analysis, spectral clustering and kernel principal component analysis are widely used in statistical learning and data mining applications. The empirical success of the kernel method is generally attributed to nonlinear feature mapping induced by the kernel, which in turn determines a low dimensional data embedding. It is important to understand the effect of a kernel and its associated kernel parameter(s) on the embedding in relation to data distributions. In this dissertation, we examine the geometry of the nonlinear embeddings for kernel PCA and kernel LDA through spectral analysis of the corresponding kernel operators. In particular, we carry out eigen-analysis of the polynomial kernel operator associated with data distributions and investigate the effect of the degree of polynomial on the data embedding. We also investigate the effect of centering kernels on the spectral property of both polynomial and Gaussian kernel operators. In addition, we extend the framework of the eigen-analysis of kernel PCA to kernel LDA by considering between-class and within-class variation operators for polynomial kernels. The results provide both insights into the geometry of nonlinear data embeddings given by kernel methods and practical guidelines for choosing an appropriate degree for dimension reduction and discrimination with polynomial kernels.

Committee: Yoonkyung Lee (Advisor); Tao Shi (Committee Member); Vincent Vu (Committee Member) Subjects: Statistics

Doctor of Philosophy, The Ohio State University, 2012, Statistics

Classification arises in a wide range of applications. A variety of statistical tools have been developed for learning classification rules from data. Understanding of their relative merits and comparisons helps users to choose a proper method in practice. This thesis focuses on theoretical comparison of model-based classification methods in statistics with algorithmic methods in machine learning in terms of the error rate. Extending Efron's comparison of logistic regression with the LDA under the normal setting, we compare classification methods based on the limiting behaviour of the classification boundary of each method. In doing so, we contrast such algorithmic methods as the support vector machine and boosting with the LDA and logistic regression and study their relative efficiencies. The analytical results also indicate some bias in the support vector machine and its variants, and we propose a proper modification for removing the bias. Besides the comparison of classification methods in efficiency, we study their robustness to model-misspecification such as non normal setting and mislabeling. In addition to the theoretical study, we also present results from numerical experiments under various settings for comparisons of finite-sample performance.

Committee: Yoonkyung Lee (Advisor); Radu Herbei (Committee Member); Tao Shi (Committee Member) Subjects: Statistics

Master of Science, The Ohio State University, 2012, Computer Science and Engineering

Sentiment analysis is a task of mining subjective information expressed in text, and has received a lot of focus from the research community in Natural Language Processing in recent years. With the rapid growth of social networks, sentiment analysis is becoming much more attractive to Natural Language Processing researchers. Identifying words or phrases that carry sentiments is a crucial task in sentiment analysis. The work in this thesis concentrates on automatically constructing polarity lexicons for sentiment analysis on social networks. One of the challenges in sentiment analysis on social networks is the lack of domain-dependent polarity lexicons and there is a need for automatically constructing sentiment lexicons for any specific domain. Two proposed methods are based on graph propagation and topic modeling. Our experiments confirm the quality of the polarity lexicons constructed using these two algorithms.

Committee: Rajiv Ramnath Professor (Advisor); Jay Ramanathan Professor (Committee Member) Subjects: Computer Science