Submitted by Irene Nascimento (irene.kessia@ufpe.br) on 2016-10-17T18:29:55Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) phd_efsb_FINAL_BIBLIOTECA.pdf: 7882904 bytes, checksum: df094c44eb4ce5be12596263047790ed (MD5) / Made available in DSpace on 2016-10-17T18:29:55Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) phd_efsb_FINAL_BIBLIOTECA.pdf: 7882904 bytes, checksum: df094c44eb4ce5be12596263047790ed (MD5) Previous issue date: 2016-08-29 / FACEPE / Context: In Software Engineering, technology transfer has been treated as a problem that concernsonly two agents (innovation and adoption agents) working together to fill the knowledge gap between them. In this scenario, the transfer is carried out in a “peer-to-peer” fashion, not changing the reality of individuals and organizations around them. This approach works well when one is just seeking the adoption of a technology by a“specific client”. However, it can not solve a common problem that is the adoption of new technologies by a large mass of potential new users. In a wider context like this, it no longer makes sense to focus on “peer-to-peer” transfer. A new way of looking at the problem is necessary. It makes more sense to approach it as diffusion of innovations, where there is an information spreading in a community, similar to that observed in epidemics. Objective: This thesis proposes a paradigm shift to show the adoption of programming languages can be formally addressed as an epidemic. This focus shift allows the dynamics of programming language adoption to be mathematically modelled as such, and besides finding models that explain the community’s behaviour when adopting programming languages, it allows some predictions to be made, helping both individuals who wish to adopt a new language that might seem to be a new industry standard, and language designers to understand in real time the adoption of a particular language by a community. Method: After a proof of concept with data from Sourceforge (2000 to 2009), data from GitHub (2009 to January 2016), a well-known open source software repository, and Stack Overflow (2008 to March 2016), a popular Q&A system for software developers, were obtained and preprocessed. Using cumulative biological growth functions, often used in epidemiological contexts, we obtained adjusted models to the data. Once with the adjusted models, we evaluated their predictive capabilities through repeated applications of hypothesis testing and statistical calculations in different versions of the models obtained after adjusting the functions to samples of different time frames from the repositories. Results: We show that programming language adoption can be formally considered an epidemiological phenomenon by adjusting a well-known mathematical function used to describe such phenomena. We also show that, using the models found, it is possible to forecast programming languages adoption. We also show that it is possible to have similar insights by observing user data, as well as data from the community itself, not using software developers as susceptible individuals. Limitations: The forecast of the adoption outcome (asymptote) needs to be taken with care because it varies depending on the sample size, which also influences the quality of forecasts in general. Unfortunately, we not always have control over the sample size, because it depends on the population under analysis. The forecast of programming language adoption is only valid for the analysed population; generalizations should be made with caution. Conclusion: Addressing programming languages adoption as an epidemiological phenomenon allows us to perform analyses not possible otherwise. We can have an overview of a population in real time regarding the use of a programming language, which allows us, as innovation agents, to adjust our technology if it is not achieving the desired “penetration”; as adoption agents, we may decide, ahead of our competitors, to adopt a seemingly promising technology that may ultimately become a standard. / Contexto: Em Engenharia de Software, transferência de tecnologia tem sido tratada como um problema pontual, um processo que diz respeito a dois agentes (os agentes de inovação e adoção) trabalhando juntos para preencher uma lacuna no conhecimento entre estes dois. Neste cenário, a transferência é realizada “ponto a ponto”, envolvendo e tendo efeito apenas nos indivíduos que participam do processo. Esta abordagem funciona bem quando se está buscando apenas a adoção da tecnologia por um “cliente” específico. No entanto, ela não consegue resolver um problema bastante comum que é a adoção de novas tecnologias por uma grande massa de potenciais novos usuários. Neste contexto mais amplo, não faz mais sentido focar em transferência ponto a ponto, faz-se necessária uma nova maneira de olhar para o problema. É mais interessante abordá-lo como difusão de inovações, onde existe um espalhamento da informação em uma comunidade, de maneira semelhante ao que se observa em epidemias. Objetivo: Esta tese de doutorado mostra que a adoção de linguagens de programação pode ser tratada formalmente como uma epidemia. Esta mudança conceitual na maneira de olhar para o fenômeno permite que a dinâmica da adoção de linguagens de programação seja modelada matematicamente como tal, e além de encontrar modelos que expliquem o comportamento da comunidade quando da adoção de uma linguagem de programação, permite que algumas previsões sejam realizadas, ajudando tanto indivíduos que desejem adotar uma nova linguagem que parece se apresentar como um novo padrão industrial, quanto ajudando projetistas de linguagens a entender em tempo real a adoção de uma determinada linguagem pela comunidade. Método: Após uma prova de conceito com dados do Sourceforge (2000 a 2009), dados do GitHub (2009 a janeiro de 2016) um repositório de projetos software de código aberto, e Stack Overflow (2008 a março de 2016) um popular sistema de perguntas e respostas para desenvolvedores de software, from obtidos e pré processados. Utilizando uma função de crescimento biológico cumulativo, frequentemente usada em contextos epidemiológicos, obtivemos modelos ajustados aos dados. Uma vez com os modelos ajustados, realizamos avaliações de sua precisão. Avaliamos suas capacidades de previsão através de repetidas aplicações de testes de hipóteses e cálculos de estatísticas em diferentes versões dos modelos, obtidas após ajustes das funções a amostras de diferentes tamanhos dos dados obtidos. Resultados: Mostramos que a adoção de linguagens de programação pode ser considerada formalmente um fenômeno epidemiológico através do ajuste de uma função matemática reconhecidamente útil para descrever tais fenômenos. Mostramos também que é possível, utilizando os modelos encontrados, realizar previsões da adoção de linguagens de programação em uma determinada comunidade. Ainda, mostramos que é possível obter conclusões semelhantes observando dados de usuários e dados da comunidade apenas, não usando desenvolvedores de software como indivíduos suscetíveis. Limitações: A previsão do limite superior da adoção (assíntota) não é confiável, variando muito dependendo do tamanho da amostra, que também influencia na qualidade das previsões em geral. Infelizmente, nem sempre teremos controle sob o tamanho da amostra, pois ela depende da população em análise. A adoção da linguagem de programação só é válida para a população em análise; generalizações devem ser realizadas com cautela. Conclusão: Abordar o fenômeno de adoção de linguagens de programação como um fenômeno epidemiológico nos permite realizar análises que não são possíveis de outro modo. Podemos ter uma visão geral de uma população em tempo real no que diz respeito ao uso de uma linguagem de programação, o que nos permite, com agentes de inovação, ajustar a tecnologia caso ela não esteja alcançando o alcance desejado; como agentes de adoção, podemos decidir por adotar uma tecnologia aparentemente promissora que pode vir a se tornar um padrão.

<p dir="ltr">Network models are essential for representing a myriad of real-world problems. Two of the most important categories of networks are centralized and distributed networks. In this thesis, we investigate the efficient resource allocation for one centralized communication network and two distributed epidemic networks.</p><p dir="ltr">In Chapter 2, we study three proposed centralized coded caching schemes with uncoded pre-fetching for scenarios where end users are grouped into classes with different file demand sets. We provide a lower bound for the transmission rate for the system with heterogeneous user profiles. Then the transmission rates of the three schemes are compared with the lower bound to evaluate their gap to optimality, and also compared with each other to show that each scheme can outperform the other two when certain conditions are met. Finally, we propose a cache distribution method that results in a minimal peak rate and a minimal average rate for one of the schemes when the users’ storage is relatively small compared with the size of the library.</p><p dir="ltr">In Chapter 3, we examine a discrete-time networked SIR (susceptible-infected-recovered) epidemic model, where the infection, graph, and recovery parameters may be time-varying. We propose a stochastic framework to estimate the system states from observed testing data and provide an analytic expression for the error of the estimation algorithm. We validate some of our assumptions for the stochastic framework with real COVID-19 testing data. We identify the system parameters with the system states from our estimation algorithm. Employing the estimated system states, we provide a novel distributed eradication strategy that guarantees at least exponential convergence to the set of healthy states. We illustrate the results via simulations over northern Indiana, USA.</p><p dir="ltr">In Chapter 4, we propose a novel discrete-time multi-virus SIR model that captures the spread of competing SIR epidemics over a population network. First, we provide a sufficient condition for the infection level of all the viruses over the networked model to converge to zero in exponential time. Second, we propose an observation model which captures the summation of all the viruses’ infection levels in each node, which represents the individuals who are infected by different viruses but share similar symptoms. We present a sufficient condition for the model to be strongly locally observable. We propose a distributed Luenberger observer for the system state estimation. We demonstrate how to calculate the observer gain for the estimator and prove that the estimation error of our proposed estimator converges to zero asymptotically with the observer gain found. We also propose a distributed feedback controller which guarantees that all viruses are eradicated at an exponential rate. We then show via simulations that the estimation error of the Luenberger observer converges to zero before the viruses die out.</p><p dir="ltr">We conclude in Chapter 5, where we summarize the findings of this thesis and introduce several challenging open research questions that arise from its results. These questions encompass a range of topics, including the design of optimal testing strategies for large populations, the investigation of estimation techniques in the presence of noisy measurement models, the extension of the SIR epidemic model to more complex models like SEIR and SAIR, and the exploration of efficient vaccine allocation schemes.</p>

Traditional Public Health decision-support can benefit from the Web and social media revolution. This dissertation presents approaches to mining social media benefiting public health epidemiology. Through discovery and analysis of trends in Influenza related blogs, a correlation to Centers for Disease Control and Prevention (CDC) influenza-like-illness patient reporting at sentinel health-care providers is verified. A second approach considers personal beliefs of vaccination in social media. A vaccine for human papillomavirus (HPV) was approved by the Food and Drug Administration in May 2006. The virus is present in nearly all cervical cancers and implicated in many throat and oral cancers. Results from automatic sentiment classification of HPV vaccination beliefs are presented which will enable more accurate prediction of the vaccine's population-level impact. Two epidemic models are introduced that embody the intimate social networks related to HPV transmission. Ultimately, aggregating these methodologies with epidemic and social network modeling facilitate effective development of strategies for targeted interventions.

Social networks play an important role in disseminating a piece of information in a population. Companies advertising a newly launched product, movie promotion, political campaigns, social awareness campaigns by governments, charity campaigns by NGOs and crowd funding campaigns by entrepreneurs are a few examples where an entity is interested in disseminating a piece of information in a target population, possibly under resource constraints. In this thesis we model information diffusion in a population using various epidemic models and study optimal campaigning strategies to maximize the reach of information. In the different problems considered in this thesis, information epidemics are modeled as the Susceptible-Infected, Susceptible-Infected-Susceptible, Susceptible-Infected-Recovered and Maki Thompson epidemic processes; however, we modify the models to incorporate the intervention made by the campaigner to enhance information propagation. Direct recruitment of individuals as spreaders and providing word-of-mouth incentives to the spreaders are considered as two intervention strategies (controls) to enhance the speed of information propagation. These controls can be implemented by placing advertisements in the mass media, announcing referral/cash back rewards for introducing friends to a product or service being advertised etc. In the different problems considered in this thesis, social contacts are modeled with varying levels of complexity---population is homogeneously mixed or follows heterogeneous mixing. The solutions to the problems which consider homogeneous mixing of individuals identify the most important periods in the campaign duration which should be allocated more resources to maximize the reach of the message, depending on the system parameters of the epidemic model (e.g., epidemics with high and low virulence). When a heterogeneous model is considered, apart from this, the solution identifies the important classes of individuals which should be allocated more resources depending upon the network considered (e.g. Erdos-Renyi, scale-free) and model parameters. These classes may be carved out based on various centrality measures in the network. If multiple strategies are available for campaigning, the solution also identifies the relative importance of the strategies depending on the network type. We study variants of the optimal campaigning problem where we optimize different objective functions. For some of the formulated problems, we discuss the existence and uniqueness of the solution. Sometimes our formulations call for novel techniques to prove the existence of a solution.