A técnica de mineração de regras de associação surgiu com o objetivo de encontrar conhecimento novo, útil e previamente desconhecido em bancos de dados transacionais, e uma grande quantidade de algoritmos de mineração de regras de associação tem sido proposta na última década. O maior e mais bem conhecido problema destes algoritmos é a geração de grandes quantidades de conjuntos freqüentes e regras de associação. Em bancos de dados geográficos o problema de mineração de regras de associação espacial aumenta significativamente. Além da grande quantidade de regras e padrões gerados a maioria são associações do domínio geográfico, e são bem conhecidas, normalmente explicitamente representadas no esquema do banco de dados. A maioria dos algoritmos de mineração de regras de associação não garantem a eliminação de dependências geográficas conhecidas a priori. O resultado é que as mesmas associações representadas nos esquemas do banco de dados são extraídas pelos algoritmos de mineração de regras de associação e apresentadas ao usuário. O problema de mineração de regras de associação espacial pode ser dividido em três etapas principais: extração dos relacionamentos espaciais, geração dos conjuntos freqüentes e geração das regras de associação. A primeira etapa é a mais custosa tanto em tempo de processamento quanto pelo esforço requerido do usuário. A segunda e terceira etapas têm sido consideradas o maior problema na mineração de regras de associação em bancos de dados transacionais e tem sido abordadas como dois problemas diferentes: “frequent pattern mining” e “association rule mining”. Dependências geográficas bem conhecidas aparecem nas três etapas do processo. Tendo como objetivo a eliminação dessas dependências na mineração de regras de associação espacial essa tese apresenta um framework com três novos métodos para mineração de regras de associação utilizando restrições semânticas como conhecimento a priori. O primeiro método reduz os dados de entrada do algoritmo, e dependências geográficas são eliminadas parcialmente sem que haja perda de informação. O segundo método elimina combinações de pares de objetos geográficos com dependências durante a geração dos conjuntos freqüentes. O terceiro método é uma nova abordagem para gerar conjuntos freqüentes não redundantes e sem dependências, gerando conjuntos freqüentes máximos. Esse método reduz consideravelmente o número final de conjuntos freqüentes, e como conseqüência, reduz o número de regras de associação espacial. / The association rule mining technique emerged with the objective to find novel, useful, and previously unknown associations from transactional databases, and a large amount of association rule mining algorithms have been proposed in the last decade. Their main drawback, which is a well known problem, is the generation of large amounts of frequent patterns and association rules. In geographic databases the problem of mining spatial association rules increases significantly. Besides the large amount of generated patterns and rules, many patterns are well known geographic domain associations, normally explicitly represented in geographic database schemas. The majority of existing algorithms do not warrant the elimination of all well known geographic dependences. The result is that the same associations represented in geographic database schemas are extracted by spatial association rule mining algorithms and presented to the user. The problem of mining spatial association rules from geographic databases requires at least three main steps: compute spatial relationships, generate frequent patterns, and extract association rules. The first step is the most effort demanding and time consuming task in the rule mining process, but has received little attention in the literature. The second and third steps have been considered the main problem in transactional association rule mining and have been addressed as two different problems: frequent pattern mining and association rule mining. Well known geographic dependences which generate well known patterns may appear in the three main steps of the spatial association rule mining process. Aiming to eliminate well known dependences and generate more interesting patterns, this thesis presents a framework with three main methods for mining frequent geographic patterns using knowledge constraints. Semantic knowledge is used to avoid the generation of patterns that are previously known as non-interesting. The first method reduces the input problem, and all well known dependences that can be eliminated without loosing information are removed in data preprocessing. The second method eliminates combinations of pairs of geographic objects with dependences, during the frequent set generation. A third method presents a new approach to generate non-redundant frequent sets, the maximal generalized frequent sets without dependences. This method reduces the number of frequent patterns very significantly, and by consequence, the number of association rules.

This diploma thesis covers KDD analysis of data and options of presentation of KDD results on Web. The paper is divided into three main sections, which follow the whole process of this thesis. In the first section are mentioned theoretical basics needed for understanding of discussed problem. In this section are described notions data matrix and domain knowledge, concept of CRISP-DM methodology, GUHA method, system LISp-Miner and implementation of GUHA method in LISp-Miner including description of core procedures 4ft-Miner and CF-Miner. The second section is dedicated to the first goal of this paper. It briefly summarizes analysis made during pre-analysis phase. Then is described process of analysis of domain knowledge in a given data set. The third part focuses on the second goal of this thesis, which is problem of presentation of KDD results on Web. This section covers brief theoretical basis for used technologies. Then is described development of export script for automatic generation of website from results found using LISp-Miner system including description of structure of the output and recommendations for work in LISp-Miner system.

Given satellite images with accompanying pixel classifications and elevation data, we propose different solutions to object detection. The first method uses hierarchical clustering for segmentation and then employs different methods of classification. One of these classification methods used domain knowledge to classify objects while the other used Support Vector Machines. Additionally, a combination of three Support Vector Machines were used in a hierarchical structure which out-performed the regular Support Vector Machine method in most of the evaluation metrics. The second approach is more conventional with different types of Convolutional Neural Networks. A segmentation network was used as well as a few detection networks and different fusions between these. The Convolutional Neural Network approach proved to be the better of the two in terms of precision and recall but the clustering approach was not far behind. This work was done using a relatively small amount of data which potentially could have impacted the results of the Machine Learning models in a negative way.

Les données portant sur des événements abondent dans notre société technologique. Une façon intéressante de présenter des données temporelles réelles pour faciliter leur interprétation est un récit généré automatiquement. La compréhension de récits implique la construction d'un réseau causal par le lecteur. Les systèmes de data-to-text narratifs semblent reconnaître l'importance des relations causales. Cependant, celles-ci jouent un rôle secondaire dans leurs planificateurs de document et leur identification repose principalement sur des connaissances du domaine. Cette thèse propose un modèle d'interprétation assistée de données temporelles par génération de récits structurés à l'aide d'un mélange de règles d'association automatiquement extraites et définies manuellement. Les associations suggèrent des hypothèses au lecteur qui peut ainsi construire plus facilement une représentation causale des événements. Ce modèle devrait être applicable à toutes les données temporelles répétitives, comprenant de préférence des actions ou activités, telles que les données d'activités de la vie quotidienne. Les règles d'association séquentielles sont choisies en fonction des critères de confiance et de signification statistique tels que mesurés dans les données d'entraînement. Les règles d'association basées sur les connaissances du monde et du domaine exploitent la similitude d'un certain aspect d'une paire d'événements ou des patrons causaux difficiles à détecter statistiquement. Pour interpréter une période à résumer déterminée, les paires d'événements pour lesquels une règle d'association s'applique sont associées et certaines associations supplémentaires sont dérivées pour former un réseau associatif. L'étape la plus importante du pipeline de génération automatique de texte (GAT) est la planification du document, comprenant la sélection des événements et la structuration du document. Pour la sélection des événements, le modèle repose sur la confiance des associations séquentielles pour sélectionner les faits les plus inhabituels. L'hypothèse est qu'un événement qui est impliqué par un autre avec une probabilité relativement élevée peut être laissé implicite dans le texte. La structure du récit est appelée le fil associatif ramifié, car il permet au lecteur de suivre les associations du début à la fin du texte. Il prend la forme d'un arbre couvrant sur le sous-réseau associatif précédemment sélectionné. Les associations qu'il contient sont sélectionnées en fonction de préférences de type d'association et de la distance temporelle relative. Le fil associatif ramifié est ensuite segmenté en paragraphes, phrases et syntagmes et les associations sont converties en relations rhétoriques. L'étape de microplanification définit des patrons lexico-syntaxiques décrivant chaque type d'événement. Lorsque deux descriptions d'événement doivent être assemblées dans la même phrase, un marqueur discursif exprimant la relation rhétorique spécifiée est employé. Un événement principal et un événement principal précédent sont déterminés pour chaque phrase. Lorsque le parent de l'événement principal dans le fil associatif n'est pas l'événement principal précédent, un anaphorique est ajouté au marqueur discursif frontal de la phrase. La réalisation de surface peut être effectuée en anglais ou en français grâce à des spécifications lexico-syntaxiques bilingues et à la bibliothèque Java SimpleNLG-EnFr. Les résultats d'une évaluation de la qualité textuelle montrent que les textes sont compréhensibles et les choix lexicaux adéquats. / Data about events abounds in our technological society. An attractive way of presenting real-life temporal data to facilitate its interpretation is an automatically generated narrative. Narrative comprehension involves the construction of a causal network by the reader. Narrative data-to-text systems seem to acknowledge causal relations as important. However, they play a secondary role in their document planners and their identification relies mostly on domain knowledge. This thesis proposes an assisted temporal data interpretation model by narrative generation in which narratives are structured with the help of a mix of automatically mined and manually defined association rules. The associations suggest causal hypotheses to the reader who can thus construct more easily a causal representation of the events. This model should be applicable to any repetitive temporal data, preferably including actions or activities, such as Activity of Daily Living (ADL) data. Sequential association rules are selected based on the criteria of confidence and statistical significance as measured in training data. World and domain knowledge association rules are based on the similarity of some aspect of a pair of events or on causal patterns difficult to detect statistically. To interpret a specific period to summarize, pairs of events for which an association rule applies are associated. Some extra associations are then derived. Together the events and associations form an associative network. The most important step of the Natural Language Generation (NLG) pipeline is document planning, comprising event selection and document structuring. For event selection, the model relies on the confidence of sequential associations to select the most unusual facts. The assumption is that an event that is implied by another one with a relatively high probability may be left implicit in the text. The structure of the narrative is called the connecting associative thread because it allows the reader to follow associations from the beginning to the end of the text. It takes the form of a spanning tree over the previously selected associative sub-network. The associations it contains are selected based on association type preferences and relative temporal distance. The connecting associative thread is then segmented into paragraphs, sentences, and phrases and the associations are translated to rhetorical relations. The microplanning step defines lexico-syntactic templates describing each event type. When two event descriptions need to be assembled in the same sentence, a discourse marker expressing the specified rhetorical relation is employed. A main event and a preceding main event are determined for each sentence. When the associative thread parent of the main event is not the preceding main event, an anaphor is added to the sentence front discourse marker. Surface realization can be performed in English or French thanks to bilingual lexico-syntactic specifications and the SimpleNLG-EnFr Java library. The results of a textual quality evaluation show that the texts are understandable and the lexical choices adequate.