Contemporary enterprises are largely dependent on Information Technology (IT), which makes decision making on IT matters important. There are numerous issues that confuse IT decision making, including contradictive business needs, financial constraints, lack of communication between business and IT stakeholders and difficulty in understanding the often heterogeneous and integrated IT systems. The discipline of IT governance aims at providing the decision making structures, processes, and relational mechanisms, needed in order for IT to support and perpetuate the business. The adjacent discipline of enterprise architecture provides a broad range of frameworks and tools for model-based management of IT. Enterprise architecture is a commonly and successfully used approach, but the frameworks need to be adapted with respect to the concerns at stake in order to become truly useful. The IT organization includes all people involved in decision making regarding IT. The quality of the IT organization differs between enterprises and depends on aspects such as: are rights and responsibilities assigned to the appropriate people, are formalized processes implemented, and does proper documentation exist? This internal IT organization efficiency is labeled IT governance maturity. One might argue that internal efficiency metrics of the IT organization are of moderate interest only. What really matters is the external effectiveness of services that the IT organization delivers to the business. This latter effectiveness is labeled IT governance performance. Even though it is reasonable to believe that enterprises with good IT governance maturity also achieve high IT governance performance, the validity of this assumption has never been tested. IT management’s ability to make well-informed decisions regarding internal IT organization matters would increase if it were possible to predict IT governance performance. The contribution of this thesis is a method for model-based IT governance decision making. The method includes a metamodel, i.e. a modeling language, and a framework for the assessment of IT governance maturity and performance. The method also allows prediction of IT governance performance.  This thesis is a composite thesis consisting of four papers and an introduction. Paper A presents an overview of the method for model-based IT governance decision making. Paper B presents the mathematical foundation of the prediction apparatus, i.e. a Bayesian network that is based on statistical data. Paper C presents how the method can be used in practice to support IT governance decision making. Finally, Paper D analyzes the correlation of IT governance maturity and performance. The analysis is based on statistical data from case studies in 35 organizations. / QC 20100909

Changes in work, along with improvements in techniques to statistically model uncertainty, have resulted in a class of groupware tools able to forecast the activities and/or attentional state of their users. This thesis represents an exploration into the design, development, and use of one such system. I describe the design and development of a groupware calendar system called Augur that is augmented with the ability to predict the attendance of its users. Using Bayesian networks, Augur models the uncertain problem of event attendance, drawing inferences based on the attributes of calendar events as well as a history of attendance provided by each user. This system was deployed to an academic workgroup and studied over the course of a semester. To more deeply explore the social implications of Augur and systems like it, I conducted a structured privacy analysis of Augur to examine the vulnerabilities inherent in this type of forecasting groupware system. I present an architecture, user interface, and probabilistic model for Augur. This work also addresses the feasibility of such a system and the challenges faced when deploying it to an academic workgroup. I also report on an exploration of the systems use by individuals, its effects on communication within working relationships, and its effectiveness with respect to the presence of domestic calendars. Finally, I present a set of implications for the workplace social environment with the introduction of Augur. Specifically, I show how the integrity of predictions generated by Augur can have consequences for the privacy of users and their representations through the shared calendar. Overall, this thesis is presented as an early exploration into the potential for a new class of forecasting groupware applications. It offers guidance and lessons learned for both designers and researchers seeking to work in this area. It also presents a complete calendar application as an example for building and studying such systems.

<p> </p><p>The international maritime industry is growing fast due to an increasing number of transportations over sea. In pace with this development, the maritime surveillance capacity must be expanded as well, in order to be able to handle the increasing numbers of hazardous cargo transports, attacks, piracy etc. In order to detect such events, anomaly detection methods and techniques can be used. Moreover, since surveillance systems process huge amounts of sensor data, anomaly detection techniques can be used to filter out or highlight interesting objects or situations to an operator. Making decisions upon large amounts of sensor data can be a challenging and demanding activity for the operator, not only due to the quantity of the data, but factors such as time pressure, high stress and uncertain information further aggravate the task. Bayesian networks can be used in order to detect anomalies in data and have, in contrast to many other opaque machine learning techniques, some important advantages. One of these advantages is the fact that it is possible for a user to understand and interpret the model, due to its graphical nature.</p><p>This thesis aims to investigate how the output from a Bayesian network can be explained to a user by first reviewing and presenting which methods exist and second, by making experiments. The experiments aim to investigate if two explanation methods can be used in order to give an explanation to the inferences made by a Bayesian network in order to support the operator’s situation awareness and decision making process when deployed in an anomaly detection problem in the maritime domain.</p><p> </p>

Even though quite different in occurrence and consequences, from a modeling perspective many natural hazards share similar properties and challenges. Their complex nature as well as lacking knowledge about their driving forces and potential effects make their analysis demanding: uncertainty about the modeling framework, inaccurate or incomplete event observations and the intrinsic randomness of the natural phenomenon add up to different interacting layers of uncertainty, which require a careful handling. Nevertheless deterministic approaches are still widely used in natural hazard assessments, holding the risk of underestimating the hazard with disastrous effects. The all-round probabilistic framework of Bayesian networks constitutes an attractive alternative. In contrast to deterministic proceedings, it treats response variables as well as explanatory variables as random variables making no difference between input and output variables. Using a graphical representation Bayesian networks encode the dependency relations between the variables in a directed acyclic graph: variables are represented as nodes and (in-)dependencies between variables as (missing) edges between the nodes. The joint distribution of all variables can thus be described by decomposing it, according to the depicted independences, into a product of local conditional probability distributions, which are defined by the parameters of the Bayesian network. In the framework of this thesis the Bayesian network approach is applied to different natural hazard domains (i.e. seismic hazard, flood damage and landslide assessments). Learning the network structure and parameters from data, Bayesian networks reveal relevant dependency relations between the included variables and help to gain knowledge about the underlying processes. The problem of Bayesian network learning is cast in a Bayesian framework, considering the network structure and parameters as random variables itself and searching for the most likely combination of both, which corresponds to the maximum a posteriori (MAP score) of their joint distribution given the observed data. Although well studied in theory the learning of Bayesian networks based on real-world data is usually not straight forward and requires an adoption of existing algorithms. Typically arising problems are the handling of continuous variables, incomplete observations and the interaction of both. Working with continuous distributions requires assumptions about the allowed families of distributions. To "let the data speak" and avoid wrong assumptions, continuous variables are instead discretized here, thus allowing for a completely data-driven and distribution-free learning. An extension of the MAP score, considering the discretization as random variable as well, is developed for an automatic multivariate discretization, that takes interactions between the variables into account. The discretization process is nested into the network learning and requires several iterations. Having to face incomplete observations on top, this may pose a computational burden. Iterative proceedings for missing value estimation become quickly infeasible. A more efficient albeit approximate method is used instead, estimating the missing values based only on the observations of variables directly interacting with the missing variable. Moreover natural hazard assessments often have a primary interest in a certain target variable. The discretization learned for this variable does not always have the required resolution for a good prediction performance. Finer resolutions for (conditional) continuous distributions are achieved with continuous approximations subsequent to the Bayesian network learning, using kernel density estimations or mixtures of truncated exponential functions. All our proceedings are completely data-driven. We thus avoid assumptions that require expert knowledge and instead provide domain independent solutions, that are applicable not only in other natural hazard assessments, but in a variety of domains struggling with uncertainties. / Obwohl Naturgefahren in ihren Ursachen, Erscheinungen und Auswirkungen grundlegend verschieden sind, teilen sie doch viele Gemeinsamkeiten und Herausforderungen, wenn es um ihre Modellierung geht. Fehlendes Wissen über die zugrunde liegenden Kräfte und deren komplexes Zusammenwirken erschweren die Wahl einer geeigneten Modellstruktur. Hinzu kommen ungenaue und unvollständige Beobachtungsdaten sowie dem Naturereignis innewohnende Zufallsprozesse. All diese verschiedenen, miteinander interagierende Aspekte von Unsicherheit erfordern eine sorgfältige Betrachtung, um fehlerhafte und verharmlosende Einschätzungen von Naturgefahren zu vermeiden. Dennoch sind deterministische Vorgehensweisen in Gefährdungsanalysen weit verbreitet. Bayessche Netze betrachten die Probleme aus wahrscheinlichkeitstheoretischer Sicht und bieten somit eine sinnvolle Alternative zu deterministischen Verfahren. Alle vom Zufall beeinflussten Größen werden hierbei als Zufallsvariablen angesehen. Die gemeinsame Wahrscheinlichkeitsverteilung aller Variablen beschreibt das Zusammenwirken der verschiedenen Einflussgrößen und die zugehörige Unsicherheit/Zufälligkeit. Die Abhängigkeitsstrukturen der Variablen können durch eine grafische Darstellung abgebildet werden. Die Variablen werden dabei als Knoten in einem Graphen/Netzwerk dargestellt und die (Un-)Abhängigkeiten zwischen den Variablen als (fehlende) Verbindungen zwischen diesen Knoten. Die dargestellten Unabhängigkeiten veranschaulichen, wie sich die gemeinsame Wahrscheinlichkeitsverteilung in ein Produkt lokaler, bedingter Wahrscheinlichkeitsverteilungen zerlegen lässt. Im Verlauf dieser Arbeit werden verschiedene Naturgefahren (Erdbeben, Hochwasser und Bergstürze) betrachtet und mit Bayesschen Netzen modelliert. Dazu wird jeweils nach der Netzwerkstruktur gesucht, welche die Abhängigkeiten der Variablen am besten beschreibt. Außerdem werden die Parameter der lokalen, bedingten Wahrscheinlichkeitsverteilungen geschätzt, um das Bayessche Netz und dessen zugehörige gemeinsame Wahrscheinlichkeitsverteilung vollständig zu bestimmen. Die Definition des Bayesschen Netzes kann auf Grundlage von Expertenwissen erfolgen oder - so wie in dieser Arbeit - anhand von Beobachtungsdaten des zu untersuchenden Naturereignisses. Die hier verwendeten Methoden wählen Netzwerkstruktur und Parameter so, dass die daraus resultierende Wahrscheinlichkeitsverteilung den beobachteten Daten eine möglichst große Wahrscheinlichkeit zuspricht. Da dieses Vorgehen keine Expertenwissen voraussetzt, ist es universell in verschiedenen Gebieten der Gefährdungsanalyse einsetzbar. Trotz umfangreicher Forschung zu diesem Thema ist das Bestimmen von Bayesschen Netzen basierend auf Beobachtungsdaten nicht ohne Schwierigkeiten. Typische Herausforderungen stellen die Handhabung stetiger Variablen und unvollständiger Datensätze dar. Beide Probleme werden in dieser Arbeit behandelt. Es werden Lösungsansätze entwickelt und in den Anwendungsbeispielen eingesetzt. Eine Kernfrage ist hierbei die Komplexität des Algorithmus. Besonders wenn sowohl stetige Variablen als auch unvollständige Datensätze in Kombination auftreten, sind effizient arbeitende Verfahren gefragt. Die hierzu in dieser Arbeit entwickelten Methoden ermöglichen die Verarbeitung von großen Datensätze mit stetigen Variablen und unvollständigen Beobachtungen und leisten damit einen wichtigen Beitrag für die wahrscheinlichkeitstheoretische Gefährdungsanalyse.

Modern society is unquestionably heavily reliant on supply of electricity. Hence, the power system is one of the important infrastructures for future growth. However, the power system of today was designed for a stable radial flow of electricity from large power plants to the customers and not for the type of changes it is presently being exposed to, like large scale integration of electric vehicles, wind power plants, residential photovoltaic systems etc. One aspect of power system control particular exposed to these changes is the design of power system control and protection functionality. Problems occur when the flow of electricity changes from a unidirectional radial flow to a bidirectional. Such an implication requires redesign of control and protection functionality as well as introduction of new information and communication technology (ICT). To make matters worse, the closer the interaction between the power system and the ICT systems the more complex the matter becomes from a reliability perspective. This problem is inherently cyber-physical, including everything from system software to power cables and transformers, rather than the traditional reliability concern of only focusing on power system components. The contribution of this thesis is a framework for reliability analysis, utilizing system modeling concepts that supports the industrial engineering issues that follow with the imple-mentation of modern substation automation systems. The framework is based on a Bayesian probabilistic analysis engine represented by Probabilistic Relational Models (PRMs) in com-bination with an Enterprise Architecture (EA) modeling formalism. The gradual development of the framework is demonstrated through a number of application scenarios based on substation automation system configurations. This thesis is a composite thesis consisting of seven papers. Paper 1 presents the framework combining EA, PRMs and Fault Tree Analysis (FTA). Paper 2 adds primary substation equipment as part of the framework. Paper 3 presents a mapping between modeling entities from the EA framework ArchiMate and substation automation system configuration objects from the IEC 61850 standard. Paper 4 introduces object definitions and relations in coherence with EA modeling formalism suitable for the purpose of the analysis framework. Paper 5 describes an extension of the analysis framework by adding logical operators to the probabilistic analysis engine. Paper 6 presents enhanced failure rates for software components by studying failure logs and an application of the framework to a utility substation automation system. Finally, Paper 7 describes the ability to utilize domain standards for coherent modeling of functions and their interrelations and an application of the framework utilizing software-tool support. / <p>QC 20140505</p>

i Abstract The purpose of this project is to compare two statistical approaches, traditional multivariate analysis and Bayesian networks, for representing the relationship between volatile compounds in kiwifruit. Compound measurements were for individual vines which were progeny of an intercross. It was expected that groupings in the data (or compounds) would give some indication of the generic nature of the biochemical pathways. Data for this project was provided by the Flavour Biotech team at Plant and Food Research. This data contained many non-detected observations which were treated as zero and to deal with them, we looked for appropriate value of c for data transformation in log(x+c). The data is ‘large p small n’ paradigm – and has much in common with data, although it is not as extreme as microarray. Principal component analysis was done to select a subset of compounds that retained most of the multivariate structure for further analysis. The reduced set of data was analyzed by Cluster analysis and Bayesian network techniques. A heat map produced by Cluster analysis and a graphical representation of Bayesian networks were presented to scientists for their comments. According to them, the two graphs complemented each other; both graphs were useful in their own unique way. Along with clusters of compounds, clusters of genotypes were represented by the heat map which showed by how much a particular compound is present in each genotype while the relation among different compounds was seen from the Bayesian networks.

Open cut coal mining in the Bowen Basin of central Queensland had disturbed in excess of 55,000 ha by the turn of the 21st century and 72,000 ha by 2006. Strong export demand in recent years (since approximately 2000) has led to greater production from existing mines and to a proliferation of new ones. Therefore, over the ensuing decades, the level of mining activity can be expected to increase substantially the areas of erstwhile agricultural land that are disturbed. As mines exhaust their resources, companies will be obliged to achieve acceptable end uses for the various domains at those sites. The possibility of having successfully rehabilitated domains at selected sites certified on a progressive basis holds some appeal. While all stakeholder groups find a return of the land to its prior use (extensive cattle-grazing) an appealing goal, mining companies walk a tight-rope. The legislation under which the early mines were established does not bind them as tightly to the environmentally friendly outcomes as applies to new mines. Nonetheless, recent legislative trends as well as companies’ own policies, encourage them to exceed society’s environmental expectations. Regardless of the end use that is designated, relinquishment is permitted only subject to a satisfactory assessment of the risks to its sustainability. Cattle-grazing is considered as a suitable end use, partly because the return of mined land to its prior use is preferred to its designation to some other use and partly because cattle could serve to reduce the bulk of pasture growth that occurs at some sites, reducing the risk of erosion if an intense fire were to occur followed by heavy rain. Graziers’ primary motivation for seeking tenure of mined land is financial. Factors that determine both a site’s productivity and its commercial ‘worthwhileness’ are examined in this research. The major focus of this research however, is the style of management that the custodial grazier may employ. Since any future custodian is likely to be a local landholder (perhaps the grazier family from whom the land was originally acquired for mining some decades earlier), it is the management style of local farmers that is of primary interest. Some graziers use the land more intensively than others: some with more sensitivity than others. Since the reconstructed landscape is inherently more fragile than undisturbed land, differences in management style could be critical to the sustainability of grazing. Factors driving, or at least being associated with, farmers’ land management decisions were identified from prior research as draft components of a risk assessment model for grazing. A survey of the characteristics and circumstances of Bowen Basin graziers was then conducted with a view to modelling their influence on graziers’ land management style. The survey ascertained the prior probabilities among the target graziers of the elements being modelled. An estimate of the role of these factors in shaping land management decisions was then obtained by eliciting the opinions of industry experts. These processes allowed development of a predictive model that estimates the likelihood of conservative and sensitive land management under various scenarios of site characteristics and grazier-based factors. Output from the model showed that the capital circumstances of a grazier’s business have an influence over the predicted management style of 25% of the difference between best-case and worst-case scenarios. There is a 17% greater likelihood of low-risk grazing where a grazier strongly wants tenure of the land for reasons that go beyond financial gain. The grazier’s underlying values and attitudes to land management account for a further 14%, followed by the operational structure of the business (12%), and the external climatic and economic environment (9%). Interventions that mining companies could implement to increase the likelihood of low-risk management has an influence of 23%. The credibility of the model’s output was evaluated by reference to real-life experiences of graziers who have managed cattle on mined land and their miner counterparts. Consistency of opinion among the consulted experts also contributed to the confidence that can be placed in the model’s findings. The model identifies the sources of risk if currently available mined land is used for grazing. It improves understanding of the situation in a holistic manner, and predicts the likelihood of low-risk grazing management under scenarios of interest to the user. The model identifies actions that mining companies could take to reduce risks associated with graziers’ management style. The model may also guide future rehabilitation work by highlighting features of rehabilitation that would make them more suited to commercially feasible and low-risk cattle production – or by purposefully and transparently planning for cattle-grazing not to be the designated end use.

[pt] Como consequência da atuação focada na eficiência da cadeia de suprimento, as organizações aumentam sua dependência nos fornecedores de matérias primas e componentes e tornam-se mais suscetíveis ao perfil de risco dos mesmos. O papel do gerenciamento de riscos em cadeias de suprimento é de entender e tentar evitar os efeitos devastadores ou mesmo os de menores amplitudes de rupturas de suprimento. Torna-se necessário para os gestores o desenvolvimento da capacidade para avaliação de risco, principalmente o risco de rupturas. É preciso estimar os riscos aceitáveis associados às possíveis rupturas e, então, desenvolver estratégias e diretrizes para o gerenciamento de tais riscos, já que, somente com o conhecimento dos mesmos, torna-se viável o desenvolvimento de planos de mitigação ou contingenciais. Nesta dissertação discutir-se-á o tema de gerenciamento e modelagem de riscos associados a fornecedores, destacando-se a aplicação de um modelo probabilístico de redes bayesianas como um recurso útil nos assuntos relacionados ao gerenciamento e à modelagem de riscos. As rupturas de suprimento serão modeladas utilizando-se um modelo probabilístico de redes bayesianas que possuem a habilidade de representar as relações de causa e efeito em um ambiente envolvendo incertezas. Um estudo de caso com a aplicação da metodologia na indústria de Refino também é apresentada. / [en] As a consequence of the supply chain management pursuit for efficiency, the organizations become more dependent on the suppliers of raw materials and more sensitive to their risk profiles. The Supply chain risk management (SCRM) objective is to understand and reduce the likelihood of disruptions. SCRM puts its effort in sharpen the notions of risk and reliability and tries to quantify them. The methodology used in this study can be used by managers to formulate supply chain risk management strategies and tactics that mitigate overall supply chain risks correlated with the most critical suppliers. This research is concerned with developing a Bayesian Network approach to model and to analyze supply chain disruptions associated with suppliers. The Bayesian Networks is a method of modeling the cause and effect of events and has proven to be a powerful tool under conditions of uncertainty. A case study is used to illustrate the application proposed to make the supply chain more reliable.

Existe uma noção intuitiva de que os custos associados a ações de suporte de projetos de gerenciamento de Tecnologia da Informação (TI), muitas vezes considerados já muito elevados e em crescimento, possuem forte vinculação com esforços empreendidos nas fases de desenvolvimento/implantação e teste. Apesar da importância de caracterizar e compreender a sistemática dessa relação, pouco tem sido feito neste domínio, principalmente devido à falta de mecanismos adequados tanto para o compartilhamento de informações entre as fases de um projeto de TI, quanto para aprender com experiências passadas. Para lidar com essa problemática, propõe-se nesta dissertação uma abordagem para estimar dinamicamente os custos de suporte de projetos de gerenciamento de TI à luz de informações provenientes das fases de desenvolvimento/implantação e teste. As estimativas de custos são calculadas a partir da integração de informações produzidas ao longo do ciclo de vida de projetos (passados). O núcleo da solução presente neste trabalho conta com um modelo Bayesiano para realizar previsão de custos de suporte, apoiado em um modelo de informação usado para persistir informações históricas. Para provar conceito e viabilidade técnica da solução proposta considerou-se, como estudo de caso, a predição de custos associados com projetos de implantação de infraestrutura de redes sem fio. Durante a avaliação é demonstrada a eficácia e eficiência do modelo, bem como discutido suas potencialidades e limitações para auxiliar no entendimento do compromisso entre custos de desenvolvimento/ implantação, teste e suporte. A avaliação conduzida fez uso de dados reais/sintéticos produzidos a partir de projetos do ISBSG e apresenta resultados próximos dos encontrados em cenários reais. Nossa abordagem obteve cerca de 80% de acerto na estimativa dos custos de suporte para os cenários avaliados. / There is an intuitive notion that the costs associated with IT management project support actions, often deemed extremely high and increasing, are directly related to the effort spent during their development/deployment and test phases. Despite the importance of systematically characterizing and understanding this relationship, little has been done in this realm mainly due to the lack of proper mechanisms for both sharing information between IT project phases and learning from past experientes. To tackle this issue, in this dissertation we proposed an approach for dynamically predicting IT management project support costs taking into account information gathered from the development/deployment and test phases. Support cost estimates are computed by integrating existing information from the lifecycle of (past) projects. The core of the solution in this work relies on a Bayesian model to perform support cost predictions, supported by an information model employed to persist historical information gathered from past projects. To prove the concept and technical feasibility of our solution we consider as a case study the prediction of costs (either development/test/support) associated with projects for the deployment of wireless network infrastructures. During the evaluation is demonstrated the effectiveness and efficiency of the model and discussed its potential and limitations in order to help understanding the trade-offs between development/deployment, test, and support costs. Our solution has been evaluated based on real/synthetics data gathered from the ISBSG dataset, and presents results similar to those found in real-life scenarios. Our solution has provided correct estimates for around 80% of the support costs for the scenarios evaluated.

Este trabalho propõe um modelo probabilístico de conhecimento e raciocínio para um agente, denominado Agente Social, cujo principal objetivo é analisar o perfil dos alunos, usuários de um Sistema Tutor Inteligente chamado AMPLIA, e compor grupos de trabalho. Para formar estes grupos, o Agente Social considera aspectos individuais do aluno e estratégias de formação de grupos. A aprendizagem colaborativa envolve relações sociais cujos processos são complexos e apresentam dificuldade para sua modelagem computacional. A fim de representar alguns elementos deste processo e de seus participantes, devem ser considerados aspectos individuais, tais como estado afetivo, questões psicológicas e cognição. Também devem ser considerados aspectos sociais, tais como a habilidade social, a aceitação e a forma em que as pessoas se relacionam e compõem seus grupos de trabalho ou estudo. Sistemas Tutores Inteligentes, Sistemas Multiagente e Computação Afetiva são áreas de pesquisa que vem sendo investigadas de forma a oferecer alternativas para representar e tratar computacionalmente alguns destes aspectos multidisciplinares que acompanham a aprendizagem individual e colaborativa. O Agente Social está inserido na sociedade de agentes do portal PortEdu que, por sua vez, fornece serviços ao ambiente de aprendizagem AMPLIA O PortEdu é um portal que provê serviços para os ambientes educacionais integrados a ele. Este portal foi modelado em uma abordagem multiagente e cada serviço oferecido é implementado por um agente específico. Os ambientes educacionais que utilizam os serviços do portal também são sociedades de agentes e, em geral, Sistemas Tutores Inteligentes. O ambiente AMPLIA (Ambiente Multiagente Probabilístico Inteligente de Aprendizagem) foi projetado para suportar o treinamento do raciocínio diagnóstico e modelagem de domínios de conhecimento incerto e complexo, como a área médica. Este ambiente usa a abordagem de Redes Bayesianas onde os alunos constróem suas próprias redes para um problema apresentado pelo sistema através de um editor gráfico de Redes Bayesianas. Neste trabalho, o editor do AMPLIA foi adaptado para uma versão colaborativa, que permite a construção das redes por vários alunos remotos conectados ao sistema. É através deste editor que o Agente Social observa e interage com os alunos sugerindo a composição dos grupos. Foram realizados experimentos práticos acompanhados por instrumentos de avaliação, com o objetivo de analisar a composição de grupos sugerida pelo Agente Social e relacioná-la com os grupos formados espontaneamente pelos alunos no ambiente de sala de aula. O resultado do trabalho individual e dos grupos também foi analisado e discutido nesta pesquisa. / This research proposes a probabilistic knowledge and reasoning model for an agent, named Social Agent, whose main goal is to analyze students' profiles and to organize them in workgroups. These students are users of an Intelligent Tutoring System named AMPLIA. In order to suggest those groups, the Social Agent considers individual aspects of the students and also strategies for group formation. Collaborative learning involves social relationships with complex processes which are difficult to model computationally. In order to represent these relationships, we should consider several aspects of the student, such as affective state, psychological issues, and cognition. We should also consider social aspects such as social ability, social acceptance and how people relate to each other, and how they compose their workgroups. Intelligent Tutoring Systems, Multiagent Systems and Affective Computing are research areas which our research group have been investigating, in order to represent and to deal computationally with multidisciplinary issues involving individual and collaborative learning. The Social Agent is part of an agent society of the PortEdu Portal, which provides services to AMPLIA. PortEdu is an educational portal which provides facilities to educational environments integrated to it. This portal has been modeled using a multiagent approach and each of its services is represented by a specific agent. The educational environments that make use of the portal's services are also agent societies and, in general, Intelligent Tutoring Systems. AMPLIA (Probabilistic Multiagent Learning Environment) has been designed in order to support diagnostic reasoning and the modeling of diagnostic hypotheses in domains with complex and uncertain knowledge, such as the medical domain. This environment uses a Bayesian Networks approach in which students build their own networks for a clinical case through a Bayesian Network graphical editor. Here, the AMPLIA editor has been adapted and extended to a collaborative version, which enables the network construction for remote students connected to the system. Through this editor, the Social Agent observes and interacts with students, suggesting the composition of workgroups. Practical experiments using assessment tools have been carried out, in order to analyze the workgroups suggested by the Social Agent and to compare them with groups naturally composed by students in the classroom. The results of the work done by individual students and by workgroups were also analyzed and discussed in this research.