Call for Exceptional Thesis LNBIP reports state-of-the-art results in areas related to business information systems and industrial application software developmenttimely, at a high level, and in both printed and electronic form. LNBIP is abstracted/indexed in ISI Proceedings, DBLP, EI and Scopus.

No abstract

The exponential growth of the Web documents has constituted the need for automatic document summarization. In this context, extractive document summarization, i.e., that task of extracting the most relevant information, removing redundancy and presenting the remained data in a coherent and cohesive structure, is a challenging task. In this paper, we propose a novel intelligent approach, namely ExDoS, that harvests benefits of both supervised and unsupervised algorithms simultaneously. To the best of our knowledge, ExDoS is the first approach to combine both supervised and unsupervised algorithms in a single framework and an interpretable manner for document summarization purpose. ExDoS iteratively minimizes the error rate of the classifier in each cluster with the help of dynamic local feature weighting. Moreover, this approach specifies the contribution of features to discriminate each class, which is a challenging issue in the summarization task. Therefore, in addition to summarizing text, ExDoS is also able to measure the importance of each feature in the summarization process. We evaluate our model both automatically (in terms of ROUGE factor) and empirically (human analysis) on the benchmark datasets: the DUC2002 and CNN/DailyMail. Results show that our model obtains higher ROUGE scores comparing to most state-of-the-art models. The human evaluation also demonstrates that our model is capable of generating informative and readable summaries.

This work is dedicated to my family. My Mother, who now deceased made me the person I am. To my wife Elena, for her understanding of my commitment to iCOP. My mentor's the Constables, Sergeants and Inspectors in the New South Wales Police Force with whom I shared a career. But in particular, to my Professor Dr Amin Beheshti, for his backbone commitment to teaching. Who believed in how I perceived technology as an enabler to policing processes and its future.Finally, I dedicate this work to the community whom I believe in and served for the past three decades. v vi Dedication

Abstract.One of the main shortcomings of the conventional classifiers is appeared when facing with datasets having multimodal distribution. To overcome this drawback, here, an efficient strategy is proposed in which a clustering phase is firstly executed over all class samples to partition the feature space into separate subspaces (clusters). Since in clustering label of samples are not considered, each cluster contains impure samples belonging to different classes. The next phase is to apply a classifier to each of the created clusters. The main advantage of this proposed distributed approach is to simplify a complex pattern recognition problem by training a specific classifier for each subspace. It is expected applying an efficient classifier to a local cluster leads to better results compared to apply it to several scattered clusters. In the validation and test phases, before make a decision about which classifier should be applied, we should find the nearest cluster to the input sample and then utilize the corresponding trained classifier. Experimental results over different UCI datasets demonstrate a significant supremacy of the proposed distributed classifier system in comparison with single classifier approaches.Keywords: Distributed classifiers, classifier ensembles, subspace classification, distributed learning, complex systems. IntroductionClassification is the most popular machine learning task in different applications.Although there are a number of different models on this major, it is still a challenge for experts to discover a model that can efficiently work on data with unknown distribution. Recently, researchers come to this agreement that a single classifier does not have enough ability, in terms of capacity and generalization, to classify a complex dataset. Among different models of classification, ensemble structures have recently attracted much attention because in many real applications they provide suitable results. As far as ensemble learners (e.g. boosting) are highly sensitive to noisy samples, this paper is aimed at proposing a completely new idea to handle datasets having multimodal density distribution. The main idea of this paper is to convert a big complex problem into some small problems with lower complexity and apply a learner to each subspace. It is obvious that the probability of getting good result by applying a classifier to a certain partition is more than expecting good results by applying just one learner for some scattered cluster of samples.