Abstract
Our society is experiencing sudden changes in work organization in part due to the growing ease with which people can collaborate. Many successful cases of peer-to-peer models of organization arise and assume leading positions in world economy replacing, in many cases, the traditional hierarchical organization. People are evolving and interacting within heterogeneous teams composed by members from many different cultural groups and with distinct skills and backgrounds. Modern economy requires engineers to excel in collaborative and communication skills at an international setting. However, these competences are not usually addressed in most engineering curricula. We believe that in such a demanding and culturally diverse environment as the labour market is today, it is essential to promote team work and communication skills at an international and intercultural level. In the Multinational Undergraduate Team Work course, MUTW, students develop their capstone project as members of an international team while working at their home institutions. MUTW projects are to be developed by teams of final-year-undergraduate students from a multinational group of higher education institutions working to solve some engineering problem. Team members are geographically spread to assure heterogeneous teams and to promote international cooperation. This paradigm can be applied in any project/internship course unit. The results from the first edition are very encouraging supporting our initial hypothesis that MUTW significantly promotes students soft skills without requiring any change to prior degree curricula. © 2011 IEEE.

Abstract
Web search engines are powerful tools used to satisfy specific information needs on the web. Their purpose is to maximize user satisfaction when performing this task. Although there are other sources of evidence, besides text, to characterize document relevance for a specific need, especially for HTML documents, current search engines do not allow users to explore these features when posing a query. Search engine queries are based almost exclusively on keywords. We believe that it is possible to improve user satisfaction if HTML tags and document metadata are available to users at query time. In this paper we present Xearch, a meta-search system that wraps public search engines in a framework that improves both the expressiveness of the language available for the user to specify information needs and the control over the answer format. Xearch converts HTML pages to a specific XML Schema, covering text and metadata derived from HTML. User queries are then submitted on this schema and can be specified through keywords but also explore documents' HTML tags and metadata. Results from our experimental evaluation confirm that it is possible to improve the answer quality with this framework.

Abstract

Abstract
The correct grading of free text answers to exam questions during an assessment process is time consuming and subject to fluctuations in the application of evaluation criteria, particularly when the number of answers is high (in the hundreds). In consequence of these fluctuations, inherent to human nature, and largely determined by emotional factors difficult to mitigate, it is natural that small discrepancies arise in the ratings assigned to similar responses. This means that two answers with similar quality may get a different grade which may generate inequities in the assessment process. Reducing the time required by the assessment process on one hand, and grouping the answers in homogenous groups, on the other hand, are the main motivations for developing the work presented here. We believe that it is possible to reduce unintentional inequities during an assessment process of free text answers by applying text mining techniques, in particular, automatic text classification, enabling to group answers in homogeneous sets comprising answers with uniform quality. Thus, instead of grading answers in random order, the teacher may assess similar answers in sequence, one after the other. The teacher may also choose, for example, to grade free text answers in decreasing order of quality, the best first, or in ascending order of quality, starting to grade the group of the worst answers. The active learning techniques we are applying throughout the grading process generate intermediary models to automatically organize the answers still not fixed in homogeneous groups. These techniques contribute to reduce the time required for the assessment process, to reduce the occurrence of grading errors and improve detection of plagiarism. © Academic Publishing International Ltd.

Abstract
In some classification tasks, such as those related to the automatic building and maintenance of text corpora, it is expensive to obtain labeled instances to train a classifier. In such circumstances it is common to have massive corpora where a few instances are labeled (typically a minority) while others are not. Semi-supervised learning techniques try to leverage the intrinsic information in unlabeled instances to improve classification models. However, these techniques assume that the labeled instances cover all the classes to learn which might not be the case. Moreover, when in the presence of an imbalanced class distribution, getting labeled instances from minority classes might be very costly, requiring extensive labeling, if queries are randomly selected. Active learning allows asking an oracle to label new instances, which are selected by criteria, aiming to reduce the labeling effort. D-Confidence is an active learning approach that is effective when in presence of imbalanced training sets. In this paper we evaluate the performance of d-Confidence in comparison to its baseline criteria over tabular and text datasets. We provide empirical evidence that d-Confidence reduces label disclosure complexity-which we have defined as the number of queries required to identify instances from all classes to learn-when in the presence of imbalanced data. © 2012 The Brazilian Computer Society.

Abstract
Collecting and annotating exemplary cases is a costly and critical task that is required in early stages of any classification process. Reducing labeling cost without degrading accuracy calls for a compromise solution which may be achieved with active learning. Common active learning approaches focus on accuracy and assume the availability of a pre-labeled set of exemplary cases covering all classes to learn. This assumption does not necessarily hold. In this paper we study the capabilities of a new active learning approach, d-Confidence, in rapidly covering the case space when compared to the traditional active learning confidence criterion, when the representativeness assumption is not met.. Experimental results also show that; d-Confidence reduces the number of queries required to achieve complete class coverage and tends to improve or maintain classification error.