No description available.

L'avènement du numérique permet de stocker de grandes quantités de parole à moindre coût. Malgré les récentes avancées en recherche documentaire audio, il reste difficile d'exploiter les documents à cause du temps nécessaire pour les écouter. Nous tentons d'atténuer cet inconvénient en produisant un résumé automatique parlé à partir des informations les plus importantes. Pour y parvenir, une méthode de résumé par extraction est appliquée au contenu parlé, transcrit et structuré automatiquement. La transcription enrichie est réalisée grâce aux outils Speeral et Alize développés au LIA. Nous complétons cette chaîne de structuration par une segmentation en phrases et une détection des entités nommées, deux caractéristiques importantes pour le résumé par extraction. La méthode de résumé proposée prend en compte les contraintes imposées par des données audio et par des interactions avec l'utilisateur. De plus, cette méthode intègre une projection dans un espace pseudo-sémantique des phrases. Les différents modules mis en place aboutissent à un démonstrateur complet facilitant l'étude des interactions avec l'utilisateur. En l'absence de données d'évaluation sur la parole, la méthode de résumé est évaluée sur le texte lors de la campagne DUC 2006. Nous simulons l'impact d'un contenu parlé en dégradant artificiellement les données de cette même campagne. Enfin, l'ensemble de la chaîne de traitement est mise en œuvre au sein d'un démonstrateur facilitant l'accès aux émissions radiophoniques de la campagne ESTER. Nous proposons, dans le cadre de ce démonstrateur, une frise chronologique interactive complémentaire au résumé parlé.

L'increment de bases de dades que cada vegada contenen imatges més difícils i amb un nombre més elevat de categories, està forçant el desenvolupament de tècniques de representació d'imatges que siguin discriminatives quan es vol treballar amb múltiples classes i d'algorismes que siguin eficients en l'aprenentatge i classificació. Aquesta tesi explora el problema de classificar les imatges segons l'objecte que contenen quan es disposa d'un gran nombre de categories. Primerament s'investiga com un sistema híbrid format per un model generatiu i un model discriminatiu pot beneficiar la tasca de classificació d'imatges on el nivell d'anotació humà sigui mínim. Per aquesta tasca introduïm un nou vocabulari utilitzant una representació densa de descriptors color-SIFT, i desprès s'investiga com els diferents paràmetres afecten la classificació final. Tot seguit es proposa un mètode par tal d'incorporar informació espacial amb el sistema híbrid, mostrant que la informació de context es de gran ajuda per la classificació d'imatges. Desprès introduïm un nou descriptor de forma que representa la imatge segons la seva forma local i la seva forma espacial, tot junt amb un kernel que incorpora aquesta informació espacial en forma piramidal. La forma es representada per un vector compacte obtenint un descriptor molt adequat per ésser utilitzat amb algorismes d'aprenentatge amb kernels. Els experiments realitzats postren que aquesta informació de forma te uns resultats semblants (i a vegades millors) als descriptors basats en aparença. També s'investiga com diferents característiques es poden combinar per ésser utilitzades en la classificació d'imatges i es mostra com el descriptor de forma proposat juntament amb un descriptor d'aparença millora substancialment la classificació. Finalment es descriu un algoritme que detecta les regions d'interès automàticament durant l'entrenament i la classificació. Això proporciona un mètode per inhibir el fons de la imatge i afegeix invariança a la posició dels objectes dins les imatges. S'ensenya que la forma i l'aparença sobre aquesta regió d'interès i utilitzant els classificadors random forests millora la classificació i el temps computacional. Es comparen els postres resultats amb resultats de la literatura utilitzant les mateixes bases de dades que els autors Aixa com els mateixos protocols d'aprenentatge i classificació. Es veu com totes les innovacions introduïdes incrementen la classificació final de les imatges. / The release of challenging data sets with ever increasing numbers of object categories isforcing the development of image representations that can cope with multiple classes andof algorithms that are efficient in training and testing. This thesis explores the problem ofclassifying images by the object they contain in the case of a large number of categories. We first investigate weather the hybrid combination of a latent generative model with a discriminative classifier is beneficial for the task of weakly supervised image classification.We introduce a novel vocabulary using dense color SIFT descriptors, and then investigate classification performances by optimizing different parameters. A new way to incorporate spatial information within the hybrid system is also proposed showing that contextual information provides a strong support for image classification. We then introduce a new shape descriptor that represents local image shape and its spatial layout, together with a spatial pyramid kernel. Shape is represented as a compactvector descriptor suitable for use in standard learning algorithms with kernels. Experimentalresults show that shape information has similar classification performances and sometimes outperforms those methods using only appearance information. We also investigate how different cues of image information can be used together. Wewill see that shape and appearance kernels may be combined and that additional informationcues increase classification performance. Finally we provide an algorithm to automatically select the regions of interest in training. This provides a method of inhibiting background clutter and adding invariance to the object instance's position. We show that shape and appearance representation over the regions of interest together with a random forest classifier which automatically selects the best cues increases on performance and speed. We compare our classification performance to that of previous methods using the authors'own datasets and testing protocols. We will see that the set of innovations introduced here lead for an impressive increase on performance.

The development of modern information technology has enabled collecting data of unprecedented size and complexity. Examples include web text data, microarray & proteomics, and data from scientific domains (e.g., meteorology). To learn from these high dimensional and complex data, traditional machine learning techniques often suffer from the curse of dimensionality and unaffordable computational cost. However, learning from large-scale high-dimensional data promises big payoffs in text mining, gene analysis, and numerous other consequential tasks. Recently developed sparse learning techniques provide us a suite of tools for understanding and exploring high dimensional data from many areas in science and engineering. By exploring sparsity, we can always learn a parsimonious and compact model which is more interpretable and computationally tractable at application time. When it is known that the underlying model is indeed sparse, sparse learning methods can provide us a more consistent model and much improved prediction performance. However, the existing methods are still insufficient for modeling complex or dynamic structures of the data, such as those evidenced in pathways of genomic data, gene regulatory network, and synonyms in text data. This thesis develops structured sparse learning methods along with scalable optimization algorithms to explore and predict high dimensional data with complex structures. In particular, we address three aspects of structured sparse learning: 1. Efficient and scalable optimization methods with fast convergence guarantees for a wide spectrum of high-dimensional learning tasks, including single or multi-task structured regression, canonical correlation analysis as well as online sparse learning. 2. Learning dynamic structures of different types of undirected graphical models, e.g., conditional Gaussian or conditional forest graphical models. 3. Demonstrating the usefulness of the proposed methods in various applications, e.g., computational genomics and spatial-temporal climatological data. In addition, we also design specialized sparse learning methods for text mining applications, including ranking and latent semantic analysis. In the last part of the thesis, we also present the future direction of the high-dimensional structured sparse learning from both computational and statistical aspects.

Made available in DSpace on 2016-08-02T21:42:57Z (GMT). No. of bitstreams: 1 Jose Claudio dos Santos Pinheiro.pdf: 5349646 bytes, checksum: 057189cedae5b7fc79c3e7cec83d51aa (MD5) Previous issue date: 2009-09-17 / This work aim to map the use of information system s theories, based on analytic resources that came from information retrieval techniques and data mining and text mining methodologies. The theories addressed by this research were Transactions Costs Economics (TCE), Resource-based view (RBV) and Institutional Theory (IT), which were chosen given their usefulness, while alternatives of approach in processes of allocation of investments and implementation of information systems. The empirical data are based on the content of textual data in abstract and review sections, of articles from ISR, MISQ and JIMS along the period from 2000 to 2008. The results stemming from the text mining technique combined with data mining were compared with the advanced search tool EBSCO and demonstrated greater efficiency in the identification of content. Articles based on three theories accounted for 10% of all articles of the three journals and the most useful publication was the 2001 and 2007.(AU) / Esta dissertação visa apresentar o mapeamento do uso das teorias de sistemas de informações, usando técnicas de recuperação de informação e metodologias de mineração de dados e textos. As teorias abordadas foram Economia de Custos de Transações (Transactions Costs Economics TCE), Visão Baseada em Recursos da Firma (Resource-Based View-RBV) e Teoria Institucional (Institutional Theory-IT), sendo escolhidas por serem teorias de grande relevância para estudos de alocação de investimentos e implementação em sistemas de informação, tendo como base de dados o conteúdo textual (em inglês) do resumo e da revisão teórica dos artigos dos periódicos Information System Research (ISR), Management Information Systems Quarterly (MISQ) e Journal of Management Information Systems (JMIS) no período de 2000 a 2008. Os resultados advindos da técnica de mineração textual aliada à mineração de dados foram comparadas com a ferramenta de busca avançada EBSCO e demonstraram uma eficiência maior na identificação de conteúdo. Os artigos fundamentados nas três teorias representaram 10% do total de artigos dos três períodicos e o período mais profícuo de publicação foi o de 2001 e 2007.(AU)

Submitted by camilla martins (camillasmmartins@gmail.com) on 2017-01-27T15:50:37Z No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) Tese_AvaliacaoAutomaticaQuestoes.pdf: 5106074 bytes, checksum: c401d50ce5e666c52948ece7af20b2c3 (MD5) / Approved for entry into archive by Edisangela Bastos (edisangela@ufpa.br) on 2017-01-30T13:02:31Z (GMT) No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) Tese_AvaliacaoAutomaticaQuestoes.pdf: 5106074 bytes, checksum: c401d50ce5e666c52948ece7af20b2c3 (MD5) / Made available in DSpace on 2017-01-30T13:02:31Z (GMT). No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) Tese_AvaliacaoAutomaticaQuestoes.pdf: 5106074 bytes, checksum: c401d50ce5e666c52948ece7af20b2c3 (MD5) Previous issue date: 2016-02-05 / Este trabalho investiga o uso de um modelo usando Latent Semantic Analysis (LSA) na avaliação automática de respostas curtas, com média de 25 a 70 palavras, de questões discursivas. Com o surgimento de ambientes virtuais de aprendizagem, pesquisas sobre correção automática tornaram-se mais relevantes, pois permitem a correção mecânica com baixo custo para questões abertas. Além disso, a correção automática permite um feedback instantâneo e elimina o trabalho de correção manual. Isto possibilita criar turmas virtuais com grande quantidade de alunos (centenas ou milhares). Pesquisas sobre avaliação automática de textos estão sendo desenvolvidas desde a década de 60, mas somente na década atual estão alcançando a acurácia necessária para uso prático em instituições de ensino. Para que os usuários finais tenham confiança, o desafio de pesquisa é desenvolver sistemas de avaliação robustos e com acurácia próxima de avaliadores humanos. Apesar de alguns estudos apontarem nesta direção, existem ainda muitos pontos a serem explorados nas pesquisas. Um ponto é a utilização de bigramas com LSA, mesmo que não contribua muito com a acurácia, contribui com a robustez, que podemos definir como confiabilidade2, pois considera a ordem das palavras dentro do texto. Buscando aperfeiçoar um modelo LSA na direção de melhorar a acurácia e aumentar a robustez trabalhamos em quatro direções: primeira, incluímos bigramas de palavras no modelo LSA; segunda, combinamos modelos de co-ocorrência de unigrama e bigramas com uso de regressão linear múltipla; terceira, acrescentamos uma etapa de ajustes sobre a pontuação do modelo LSA baseados no número de palavras das respostas avaliadas; quarta, realizamos uma análise da distribuição das pontuações atribuídas pelo modelo LSA contra avaliadores humanos. Para avaliar os resultados comparamos a acurácia do sistema contra a acurácia de avaliadores humanos verificando o quanto o sistema se aproxima de um avaliador humano. Utilizamos um modelo LSA com cinco etapas: 1) pré- processamento, 2) ponderação, 3) decomposição a valores singulares, 4) classificação e 5) ajustes do modelo. Para cada etapa explorou-se estratégias alternativas que influenciaram na acurácia final. Nos experimentos obtivemos uma acurácia de 84,94% numa avaliação comparativa contra especialistas humanos, onde a correlação da acurácia entre especialistas humanos foi de 84,93%. No domínio estudado, a tecnologia de avaliação automática teve resultados próximos aos dos avaliadores humanos mostrando que esta alcançando um grau de maturidade para ser utilizada em sistemas de avaliação automática em ambientes virtuais de aprendizagem. / This work investigates the use of a model using Latent Semantic Analysis (LSA) In the automatic evaluation of short answers, with an average of 25 to 70 words, of questions Discursive With the emergence of virtual learning environments, research on Automatic correction have become more relevant as they allow the mechanical correction With low cost for open questions. In addition, automatic Feedback and eliminates manual correction work. This allows you to create classes With large numbers of students (hundreds or thousands). Evaluation research Texts have been developed since the 1960s, but only in the The current decade are achieving the necessary accuracy for practical use in teaching. For end users to have confidence, the research challenge is to develop Evaluation systems that are robust and close to human evaluators. despite Some studies point in this direction, there are still many points to be explored In the surveys. One point is the use of bigrasms with LSA, even if it does not contribute Very much with the accuracy, contributes with the robustness, that we can define as reliability2, Because it considers the order of words within the text. Seeking to perfect an LSA model In the direction of improving accuracy and increasing robustness we work in four directions: First, we include word bigrasms in the LSA model; Second, we combine models Co-occurrence of unigram and bigrams using multiple linear regression; third, We added a stage of adjustments on the LSA model score based on the Number of words of the responses evaluated; Fourth, we performed an analysis of the Of the scores attributed by the LSA model against human evaluators. To evaluate the We compared the accuracy of the system against the accuracy of human evaluators Verifying how close the system is to a human evaluator. We use a LSA model with five steps: 1) pre-processing, 2) weighting, 3) decomposition a Singular values, 4) classification and 5) model adjustments. For each stage it was explored Strategies that influenced the final accuracy. In the experiments we obtained An 84.94% accuracy in a comparative assessment against human Correlation among human specialists was 84.93%. In the field studied, the Evaluation technology had results close to those of the human evaluators Showing that it is reaching a degree of maturity to be used in Assessment in virtual learning environments. Google Tradutor para empresas:Google Toolkit de tradução para appsTradutor de sitesGlobal Market Finder.

The work deals with automatic summarization of documents in HTML format. As a language of web documents, Czech language has been chosen. The project is focused on algorithms of text summarization. The work also includes document preprocessing for summarization and conversion of text into representation suitable for summarization algorithms. General text mining is also briefly discussed but the project is mainly focused on the automatic document summarization. Two simple summarization algorithms are introduced. Then, the main attention is paid to an advanced algorithm that uses latent semantic analysis. Result of the work is a design and implementation of summarization module for Python language. Final part of the work contains evaluation of summaries generated by implemented summarization methods and their subjective comparison of the author.

Recommender systems provide new opportunities for retrieving personalized information on the Internet. Due to the availability of big data, the fitness industries are now focusing on building an efficient recommender system for their end-users. This thesis investigates the possibilities of building an efficient recommender system for gym users. BRP Systems AB has provided the gym data for evaluation and it consists of approximately 896,000 customer interactions with 8 features. Four different matrix factorization methods, Latent semantic analysis using Singular value decomposition, Alternating least square, Bayesian personalized ranking, and Logistic matrix factorization that are based on implicit feedback are applied for the given data. These methods decompose the implicit data matrix of user-gym group activity interactions into the product of two lower-dimensional matrices. They are used to calculate the similarities between the user and activity interactions and based on the score, the top-k recommendations are provided. These methods are evaluated by the ranking metrics such as Precision@k, Mean average precision (MAP) @k, Area under the curve (AUC) score, and Normalized discounted cumulative gain (NDCG) @k. The qualitative analysis is also performed to evaluate the results of the recommendations. For this specific dataset, it is found that the optimal method is the Alternating least square method which achieved around 90\% AUC for the overall system and managed to give personalized recommendations to the users.