O reconhecimento de autoria é uma área de pesquisa efervescente, com muitas aplicações, incluindo detecção de plágio, análise de textos históricos, reconhecimento de mensagens terroristas ou falsificação de documentos. Modelos teóricos de redes complexas já são usados para o reconhecimento de autoria, mas alguns aspectos importantes têm sido ignorados. Neste trabalho, exploramos a dinâmica de redes de co-ocorrência e a relação com as palavras que representam os nós e descobrimos que ambas são claras assinaturas de autoria. Com otimização dos descritores da topologia das redes e de algoritmos de aprendizado de máquina, foi possível obter taxas de acerto maiores que 85%, sendo atingida uma taxa de 98.75% em um caso específico, para coleções de 80 livros, cada uma compilada de 8 autores de língua inglesa com 10 livros por autor. Esta tese demonstra que existem ainda aspectos inexplorados das redes de co-ocorrência de textos, o que deve permitir avanços ainda maiores no futuro próximo. / Authorship attribution is an active research area with many applications, including detection of plagiarism, analysis of historical texts, terrorist message identification or document falsification. Theoretical models of complex networks are already used for authorship attribution, but some issues have been ignored. In this thesis, we explore the dynamics of co-occurrence networks and the role of words, and found that they are both clear signatures of authorship. Using optimized descriptors for the network topology and machine learning algorithms, it has been possible to achieve accuracy rates above 85%, with a rate of 98.75% being reached in a particular case, for collections of 80 books produced by 8 English-speaking writers with 10 books per author. It is also shown that there are still many unexplored aspects of co-occurrence networks of texts, which seems promising for near future developments.

This thesis project addresses the machine learning (ML) modelling aspects of the problem of automatically extracting typological linguistic information of natural languages spoken in South Asia from annotated descriptive grammars. Without getting stuck into the theory and methods of Natural Language Processing (NLP), the focus has been to develop and test a machine learning (ML) model dedicated to the information extraction part. Starting with the existing state-of-the-art frameworks to get labelled training data through the structured representation of the descriptive grammars, the problem has been modelled as a supervised ML classification task where the annotated text is provided as input and the objective is to classify the input to one of the pre-learned labels. The approach has been to systematically explore the data to develop understanding of the problem domain and then evaluate a set of four potential ML algorithms using predetermined performance metrics namely: accuracy, recall, precision and f-score. It turned out that the problem splits up into two independent classification tasks: binary classification task and multiclass classification task. The four selected algorithms: Decision Trees, Naïve Bayes, Support VectorMachines, and Logistic Regression belonging to both linear and non-linear families ofML models are independently trained and compared for both classification tasks. Using stratified 10-fold cross validation performance metrics are measured and the candidate algorithms  are compared. Logistic Regression provided overall best results with DecisionTree as the close follow up. Finally, the Logistic Regression model was selected for further fine tuning and used in a web demo for typological information extraction tool developed to show the usability of the ML model in the field.

O reconhecimento de autoria é uma área de pesquisa efervescente, com muitas aplicações, incluindo detecção de plágio, análise de textos históricos, reconhecimento de mensagens terroristas ou falsificação de documentos. Modelos teóricos de redes complexas já são usados para o reconhecimento de autoria, mas alguns aspectos importantes têm sido ignorados. Neste trabalho, exploramos a dinâmica de redes de co-ocorrência e a relação com as palavras que representam os nós e descobrimos que ambas são claras assinaturas de autoria. Com otimização dos descritores da topologia das redes e de algoritmos de aprendizado de máquina, foi possível obter taxas de acerto maiores que 85%, sendo atingida uma taxa de 98.75% em um caso específico, para coleções de 80 livros, cada uma compilada de 8 autores de língua inglesa com 10 livros por autor. Esta tese demonstra que existem ainda aspectos inexplorados das redes de co-ocorrência de textos, o que deve permitir avanços ainda maiores no futuro próximo. / Authorship attribution is an active research area with many applications, including detection of plagiarism, analysis of historical texts, terrorist message identification or document falsification. Theoretical models of complex networks are already used for authorship attribution, but some issues have been ignored. In this thesis, we explore the dynamics of co-occurrence networks and the role of words, and found that they are both clear signatures of authorship. Using optimized descriptors for the network topology and machine learning algorithms, it has been possible to achieve accuracy rates above 85%, with a rate of 98.75% being reached in a particular case, for collections of 80 books produced by 8 English-speaking writers with 10 books per author. It is also shown that there are still many unexplored aspects of co-occurrence networks of texts, which seems promising for near future developments.