http://hdl.handle.net/2099.1/5401 2013-05-24T01:51:21Z http://hdl.handle.net/2099.1/17175 Title: Towards human interaction analysis Authors: Gavari, Maedeh Aghaei Abstract: Modeling and recognizing human behaviors in a visual surveillance task is receiving increasing attention from computer vision and machine learning researchers. Such a system should deal in particularly with detecting when interactions between people occur and classifying the type of interaction. In this work we study a flexible model for detecting human interactions. This has been done by detecting the people in the scene and retrieving their corresponding pose and position sequentially in each frame of the video. To achieve this goal our work relies on robust object detection algorithm which is based on discriminatively trained part based models to detect the human bodies in videos. We apply a ‘Gaussian Mixture Models based’ method for background subtraction and human segmentation. The output from the segmentation method which is labeled human body is combined with the background subtraction methods to obtain a bounding box around each person in images to improve the task of human body pose detection. To gain more precise pose detection models, we trained the algorithm on large, challenging but reliable dataset (PASCAL 2010). Our method is applied in home-made database comprising depth data from Kinect sensors. After successfully getting in every image sequence the corresponding label for each person as well as their pose and position, understanding of human motion comes naturally which is an important step towards human interaction analysis. 2013-02-11T00:00:00Z http://hdl.handle.net/2099.1/17174 Title: Robust Part of Speech Tagging Authors: Martínez Garcia, Eva Description: Generally, NLP tools use well-formed and annotated data to learn patterns by using machine learning techniques. However, in this work we will focus on the language used in an on-line platform for machine translation. In this area it is usual to have a framework such the following: a web-page which offer a service of translation between pairs of languages. The problem is that the casual users utilize the service to translate any type of text (cut and paste, single words, bad formatting, snipets, informal language, pre-traductions, etc.). Hence, in this situation we will find very often words with mistakes that make the system provides a bad translation because it is not able to understand the input.; The main goal of our work is, once we have identified the problem of dealing with non-standard-input is to develop a robust PoS tagger from the SVMTagger. 2013-02-11T00:00:00Z http://hdl.handle.net/2099.1/17173 Title: Bayesian Gaussian network classifiers for mass spectra classification Authors: Bellón Molina, Víctor Manuel Description: The early diagnosis of diseases in patients is a key objective of biomedical science and one of the most important factors in the treatment of diseases such as cancer. The early detection of cancer can make the di erence between a successful treatment and the dead of the patient. Ovarian cancer is diagnosed at late clinical stage in more than 80% of patients and the 5-year survival rate is around 35% of population, while in early diagnosed patients it exceeds 90%. The aim of this work is to present techniques for the early detection of ovarian cancers based in probabilistic analysis of proteomic spectra. 2013-02-11T00:00:00Z http://hdl.handle.net/2099.1/16512 Title: Detecting malicious profiles in Twitter Authors: Díaz, Vicente Abstract: The popularity of Social Networks during the last years has caught the attention of cybercriminals for the distribution of Spam and malicious contents. In order to do that, they create fake profiles to send spam messages to legitimate users, leading to fraud or malware campaigns. Sometimes cybercriminals use stolen accounts of legitimate users to send these malicious messages. The goal of this work is to use information available for any user to detect malicious profiles in Twitter, the second most popular Social Network in the world. Also we explore the possibility of distinguishing into different kind of malicious profiles: spammers and hacked accounts. We show how it is possible to obtain a set of features derived from the public information available in order to correctly classify malicious and clean profiles with a success rate over 90%. We show, also, how the same method could be used to detect hacked profiles with similar results. Based on these results, we propose a global system that could use both local and global information for improved results in the detection of malicious profiles. 2012-11-09T00:00:00Z http://hdl.handle.net/2099.1/16505 Title: Automatic classification of attention-deficit/hyperactivity disorder using brain activation Authors: Vila Muñoz, Maria del Mar Abstract: Nowadays, there is an active fi eld of research in neuroscience trying to fi nd relations between neurofunctional abnormalities of brain structures and neurological disorders. Previous statistical studies on brain functional Magnetic Resonance Images (fMRI) have found Attention Defi cit Hyperactivity Disorder (ADHD) patients are characterized by reduced activity in the inferior frontal gyrus (IFG) during response inhibition tasks and in the Ventral Striatum (VStr) during reward anticipation tasks. Interpreting brain image experiments using fMRI requires analysis of complex data and diff erent univariate or multivariate approaches can be chosen. Recently, one analysis approach that has grown in popularity is the use of machine learning algorithms to train classifiers to discriminate abnormal behavior or other variables of interest from fMRI data. The purpose of this work is to apply machine learning techniques to perform fMRI group analysis in an adult population. We propose a multivariate classifi er using diff erent discriminative features. Furthermore, we show how temporal information of fMRI data can be taken into account to improve the discrimination. We demonstrate that our new approach is able to diagnose the ADHD characteristics based on the activation in the executive functions. Previous results on the response inhibition task did not find di fferences between activation responses. Opposite to these results, we achieve accurate classifi cation performance for this task. Moreover, in this case, we show that classi fication rates can be signi cantly improved by incorporating temporal information into the classi fier. 2012-11-08T00:00:00Z http://hdl.handle.net/2099.1/16504 Title: Automatic demarcation for videofluoroscopy swallowing study Authors: Lizana García, Matias Abstract: Videofluoroscopy tests are designed to analyze the swallowing response of the patient. Lot of patients die due to swallowing disorders, and physicians want to analyze this procedure to detect this swallowing problems. The principal problem is that the time spent to analyze a video is so long, and has to be done manually, and this derives to a high cost of time and money to the health system. Hyoid Marker is an application that provides an automatic demarcation for videofluoroscopy studies, saving to the physician the time to mark important objects in the video, and allowing him to dedicate his time analyzing only the medical aspects of the video. 2012-11-08T00:00:00Z http://hdl.handle.net/2099.1/16496 Title: Opinion mining from a large corpora of natural language reviews Authors: Fiz Pontiveros, Beltrán Borja Abstract: This master thesis is focused on the development of a system for automatically processing a large database of textual hotel reviews in natural language to extract relevant opinions from users on a series of predefined features of quality (service, food, location, etc) The information extracted has to be categorized according to polarity (positive/negative opinions) and arranged so that the final search application can use it to display complementary information of each hotel based on the extracted opinions. Initially a set of hotel reviews is data mined from online sources; a subset of this dataset is then filtered and manually annotated to create a corpus and to help with the creation of a taxonomy for the domain of hotel reviews. A system is then designed to detect, extract and evaluate opinions, and evaluated using the corpus built. 2012-11-07T00:00:00Z http://hdl.handle.net/2099.1/16495 Title: A study of feature selection algorithms for accuracy estimation Authors: Butt, Kashif Javed Abstract: The main purpose of Feature Subset Selection is to find a reduced subset of attributes from a data set described by a feature set. The task of a feature selection algorithm (FSA) is to provide with a computational solution motivated by a certain defi nition of relevance or by a reliable evaluation measure. Feature weighting is a technique used to approximate the optimal degree of influence of individual features using a training set. When successfully applied relevant features are attributed a high weight value, whereas irrelevant features are given a weight value close to zero. Feature weighting can be used not only to improve classi cation accuracy but also to discard features with weights below a certain threshold value and thereby increase the resource efi ciency of the classifier. In this work several fundamental feature weighting algorithms (FWAs) are studied to assess their performance in a controlled experimental scenario. A measure to evaluate FWAs score is devised that computes the degree of matching between the output given by a FWAs and the known optimal solutions. A study of relation between the score obtained from the di fferent classi fiers, variance of the score in the di fferent sample size is carried out as well as the relation between the score and the estimated probability of error of the model (Pe) for the classification problems and the square error (e2) for the regression problem. 2012-11-07T00:00:00Z http://hdl.handle.net/2099.1/16449 Title: Data analysis and navigation in high-dimensional chemical and biological spaces Authors: Cester Bofarull, Josep Description: The goal of this master thesis is to develop and validate a visual data-mining approach suitable for the screening of chemicals in the context of REACH [Registration, Evaluation, Authorization and Restriction of Chemicals]. The proposed approach will facilitate the development and validation of non-testing methods via the exploration of environmental endpoints and their relationship with the chemical structure and physicochemical properties of chemicals. The use of an interactive chemical space data exploration tool using 3D visualization and navigation will enrich the information available with additional variables like size, texture and color of the objects of the scene (compounds). The features that distinguish this approach and make it unique are (i) the integration of multiple data sources allowing the recovery in real time of complementary information of the studied compounds, (ii) the integration of several algorithms for the data analysis (dimensional reduction, generation of composite variables and clustering) and (iii) direct user interaction with the data through the virtual navigation mechanism. All this is achieved without the need for specialized hardware or the use of specific devices and high-cost virtual reality and mixed reality. 2012-10-31T00:00:00Z http://hdl.handle.net/2099.1/16446 Title: Underwater signals classification and clustering Authors: Martí Farriol, Jaume Abstract: This thesis deals with the problem of classification and clustering of tonal sounds. Tonal sounds are a class of sounds that are localized in frequency and extend over a time with frequency modulation. To achieve those goals, the problem of tonal extraction is also studied. Description: The presence of noise in the oceans has been growing in the recent years due to the increasing levels of human activity on the seas. Those activities include the off shore industry, petrol industry and transport industries. This noise contamination damages the marine ecosystem. In order to plan human activities on the seas in such a way that they produce the minimum alteration to the marine ecosystem it is very beneficial to be able to understand the relation between anthropogenic and natural process. One the practical side, this thesis could be the starting point for future technologies that could help plan human activities in such a way that mitigate the most the impact they produce on the ocean environment. A part from protection of the ocean fauna from hearing loss and damage to their sonars this project could also help researchers to monitor marine fauna. (...) The aim of this thesis is the classifi cation in real time of underwater frequency modulated acoustic signals into predefi ned groups. The signals are constrained into those being frequency modulated because the automatic classi cation of impulsive sounds has already been subject of considerable study. Apart from classi cation, clustering of tonal sounds is also a topic of this thesis. The kind of acoustic emissions that can be found on the underwater environment in the seas and oceans of the planet come from 3 broad categories, those categories are biological sources, i.e. dolphins, baleens, etc; anthropogenic sources, i.e. ship motors, ship sonars, etc; and natural phenomena like earthquakes and rain. Of those categories, those that produce frequency modulated tonal sounds are marine mammals and some ship sonars. The underwater signals studied in this thesis are obtained from deep sea platforms and o ffshore moored stations, and are transmitted wirelessly or though a cord to coastal computer centers, where the signals are analyzed. 2012-10-31T00:00:00Z http://hdl.handle.net/2099.1/16445 Title: Automatic segmentation of Nucleus Accumbens Authors: Sem, Federico Abstract: Segmentation of subcortical structures in the brain has become an increasingly important topic in contemporary medicine. The ability to effi ciently isolate different regions of the human brain has allowed doctors and technicians to become more e fficient in the diagnosis of mental disorders and the evaluation of the patient conditions. An area of the brain whose possible segmentation has received particular attention is the Nucleus Accumbens, which is believed to play a central role in the reward circuit. In fact, studies of volumetric brain magnetic resonance imaging (MRI) have shown neuroanatomical abnormalities of this structure in adult attention defficit/hyperactivity disorder (ADHD), and speci cally a smaller average volume of the region. The use of a reliable automated segmentation method would therefore represent an extremely helpful and e fficient tool for identifying this disorder, especially when compared to manual volume labeling methods, which often turn out to be tedious and extremely time-consuming. However, automatic segmentation of the Accumbens is extremely di fficult to obtain, due to the lack of contrast with the surrounding structures. This means that most conventional segmentation methods are useless for this purpose, and makes the segmentation method selection a very delicate procedure. Consequently, the main objective of the thesis is the implementation of a robust algorithm for segmenting the Nucleus Accumbens structure. The research project aims to apply pre-existing segmentation methods to the Nucleus Accumbens, moving then to an evaluation of such methods and an estimation of how e ffective they are. Diff erent segmentation methods were used for this purpose; firstly, the standard Atlas Segmentation Approach was used, showing generally poor results paired with long computational times and high complexity. Moreover, this method has shown potential problems in the individuation of the correct region, leading, in some cases, to completely wrong segmentations. In addition to the fi rst method, Multi Atlas Segmentation and Adaptive Multi Atlas Segmentation methods have been implemented. The results have shown improved accuracy and better performance than the original method. Judging by the results, the segmentation of the Nucleus Accumbens has proven to be an extremely complicated task, both for the dimension of the structure itself and for the lack of contrast with the surrounding structures. In order to improve detection accuracy, combination of multiple methods is necessary, as using a single method for the segmentation process can lead to an incorrect labeling. 2012-10-31T00:00:00Z http://hdl.handle.net/2099.1/16444 Title: Development of a tourism recommender system Authors: Ciurana Simó, Emili Roger Abstract: Nowadays, many people rely on online services to plan a trip. However, they are usually faced with the problem of being supplied with lots of information. In consequence, they have to invest a great deal of time to decide what to visit, when, etc. This huge amount of possibilities available on the net makes it difficult for users to discern the more interesting offers from the rest. As a result, the more appealing offers can go unnoticed. In order to improve the tourist experience, recommender systems offer personalised information to users. In other words, the system selects the more suitable and adequate offers for users and offers activities appropriate to their profile. In this thesis, we present the EnoSigTur system, a smart recommender system for tourists interested in experiences related to the wine sector. It has been developed in the Parc Científic i Tecnològic de Turisme i Oci of Vila-seca with the collaboration of researchers in the Universitat Rovira i Virgili. Trough a web application, the system allows users to know wine production activities available in the region of Tarragona. Users just have to indicate their interests in general terms and the system will select the more convenient activities for them. EnoSigTur is capable of modifying the initial information of the users preferences by studying the interaction between the user and the system, and offering them more adjusted recommendations. This system also allows users to plan a trip by providing advanced planning services; for example, date, length of the trip, etc. A mobile phone application will permit users to monitor the planned trip while it is taking place. EnoSigTur is designed to supply a user-friendly and flexible service either to visitors with a superficial knowledge of the wine production area, or to experienced visitors who have already been in contact with these types of activities. Moreover, as we suggested before, it provides personalised recommendations according to users interests, gives clients the necessary tools to plan the trip and makes it possible for them to discover other activities in the region. In the following chapters we will deal with the main problems that tourism presents in terms of information search and decision-making processes. We will also present the recommender systems and the ontology, so that the reader will be able to grasp the gist of the project. Finally, we will give details of our recommender system. 2012-10-31T00:00:00Z http://hdl.handle.net/2099.1/16443 Title: Dynamic learning and refinement of preferences through keywords Authors: Perelló Cruz, David Description: The main goal in this work is to learn user preferences in situations where the objects to be treated are formed only by textual information and we continuously have information of selections made by the user. This work has been divided in two major parts: the first one including the algorithms and techniques to rank a set of alternatives, and the second one including the techniques to maintain the profile up to date. Regarding the first part, the goal is to evaluate an object of type text, i.e. given the user preferences to assign the degree of potential interest on that object. This will allow us to evaluate the set of alternatives and to sort them according to the user preferences. Concerning the second part, the main goal is to design a method to update the user profile, given the user selection from a set of alternatives in the first part. This method will allow to adapt a user profile in an unsupervised and dynamic way. To achieve these objectives it is necessary to fulfil the tasks discussed in this document and named below in the document organization. 2012-10-31T00:00:00Z http://hdl.handle.net/2099.1/16438 Title: Exploratory data analysis using network based techniques Authors: Granell Martorell, Clara Description: The aim of this document is to present the work done during the development of my master thesis. The work belongs to the field of complex networks, more concretely to the detection of communities in complex networks. Chapter 1 will be an introduction of the basic concepts and motivations of this work, mainly clarifying the fields of exploratory data analysis, data clustering and complex networks. As all the work is about the finding of communities in complex networks, Chapter 2 is devoted to explain the concepts of mesoscopic structure of networks and its importance in the analysis of real networks, along with the explanations of some of the most well-known techniques to perform this analysis. All the progress done during the master thesis relies on a method for detecting communities developed in the past years by the research group I belong to. This method is known as the AFG algorithm, named after the three authors Arenas, Fernández and Gómez, and it is explained in section 2.5.2 with special emphasis. The work that I have developed is composed of two separate problems: the first one consists in designing an application to make possible the use of the AFG community detection method to perform data clustering over real world multidimensional datasets, which is explained in Chapter 3. The second work consists in improving the AFG method to make possible the detection of communities even when the difference of sizes of the communities make their detection impossible for other community detection algorithms, which can be found in Chapter 4. Chapter 5 contains the conclusions and the future lines of research derived from the present work, and in the Appendix there is a list of publications that sustain the contents presented in this document. 2012-10-30T00:00:00Z http://hdl.handle.net/2099.1/16436 Title: Moving objects and removing obstacles with two robotic hands Authors: Rodr guez Pacheco, Carlos Arturo Abstract: This work deals with the problem of planning the movements of a two-hand system in order to grasp an object with one hand and using the other to remove potential obstacles. The approach is based on a Probabilistic Road Map that does not rule out samples with collisions with removable objects but instead classifies them according to the collided obstacle(s), and allows the search of free paths with the indication of which objects must be removed from the work-space to make the path be valid. The approach has been implemented and some examples are presented in this work. 2012-10-30T00:00:00Z