Eduardo N. Borges

An unsupervised heuristic-based approach for bibliographic metadata deduplication

Digital libraries of scientific articles contain collections of digital objects that are usually described by bibliographic metadata records. These records can be acquired from different sources and be represented using several metadata standards. These metadata standards may be heterogeneous in both, content and structure. All of this implies that many records may be duplicated in the repository, thus affecting the quality of services, such as searching and browsing. In this article we present an approach that identifies duplicated bibliographic metadata records in an efficient and effective way. We propose similarity functions especially designed for the digital library domain and experimentally evaluate them. Our results show that the proposed functions improve the quality of metadata deduplication up to 188% compared to four different baselines. We also show that our approach achieves statistical equivalent results when compared to a state-of-the-art method for replica identification based on genetic programming, without the burden and cost of any training process.

Consensus via penalty functions for decision making in ensembles in fuzzy rule-based classification systems

Display Omitted A consensus method via penalty functions for decision making in ensembles of fuzzy rule-based classification systems is introduced.Overlap indices are built using overlap functions.A method for constructing confidence and support measures from overlap indices is presented.A new fuzzy rule mechanism is proposed, considering different overlap indices, which generalizes the classical methods.An example of a generation of fuzzy rule-based ensembles and the decision making by consensus via penalty functions is presented. The aim of this paper is to propose a consensus method via penalty functions for decision making in ensembles of fuzzy rule-based classification systems (FRBCSs). For that, we first introduce a method based on overlap indices for building confidence and support measures, which are usually used to evaluate the degree of certainty or interest of a certain association rule. Those overlap indices (a generalizations of the Zadehs consistency index between two fuzzy sets) are built using overlap functions, which are a special kind of non necessarily associative aggregation functions proposed for applications related to the overlap problem and/or when the associativity property is not demanded. Then, we introduce a new FRM for the FRBCS, considering different overlap indices, which generalizes the classical methods. By considering several overlap indices and aggregation functions, we generate fuzzy rule-based ensembles, providing different results. For the decision making related to the selection of the best class, we introduce a consensus method for classification, based on penalty functions. We also present theoretical results related to the developed methods. A detailed example of a generation of fuzzy rule-based ensembles based on the proposed approach, and the decision making by consensus via penalty functions, is presented.

Using the Choquet Integral in the Pooling Layer in Deep Learning Networks

This paper aims to introduce the proposal of replacing the usual pooling functions by the Choquet integral in Deep Learning Networks. The Choquet integral is an aggregation function studied and applied in several areas, as, e.g., in classification problems. Its importance is related to the fact that it considers the relationship between the data to be aggregated by means of a fuzzy measure, unlike other aggregation functions such as the arithmetic mean and the maximum. The idea of this paper is to use the Choquet integral to reduce the size of an image, obtaining an abstract form of representation, that is, reducing the perception of the network corresponding to small changes in the image. The use of this aggregation function in the place of the max-pooling and mean-pooling functions of Convolutional Neural Networks presented promising results. This assertion is based on the Normalized Cross-Correlation and Structural Content quality measures applied to the original images and resulting images. It is important to emphasize that this preliminary study of Choquet integral as a pool layer has not yet been implemented on Convolutional Neural Networks until the present moment.

The effects of classifiers diversity on the accuracy of stacking.

In recent years several data classification techniques have been proposed. However, it is not a trivial task to choose the most appropriate classifier for deal with a particular problem and set it up properly. In addition, there is no optimal algorithm to solve all prediction problems. In order to improve the result of the classification process, the stacking strategy combines the knowledge acquired by individual learning algorithms aiming to discover new patterns not yet identified. Stacking combines the outputs of base classifiers, induced by several learning algorithms using the same dataset, by means of a meta-classifier. The main goal of this paper is to evaluate the effects of classifier diversity on the accuracy of stacking. We have performed a lot of experiments which results show the impact of multiple diversity measures on the gain of stacking, considering many real datasets extracted from UCI machine learning repository and three synthetic twodimensional datasets. The results revealed connections between some measures and the gain of stacking, but they imply a weak or moderate relationship that suggest predicting the improvement on the best base classifier accuracy using diversity measures is inappropriate.

Contact Deduplication in Mobile Devices using Textual Similarity and Machine Learning.

This paper presents a method that identifies duplicate contacts, i.e., records representing the same person or organization, automatically collected from multiple data sources. Contacts are compared using similarity functions, which scores are combined by a classification model based on decision trees, avoiding the need for an expert to manually configure similarity thresholds. The experiments show that the proposed method identified correctly up to 92% of duplicate contacts.

An Analysis of the Impact of Diversity on Stacking Supervised Classifiers.

Due to the growth of research in pattern recognition area, the limits of the techniques used for the classification task are increasingly tested. Thus, it is clear that specialized and properly configured classifiers are quite effective. However, it is not a trivial task to choose the most appropriate classifier for deal with a particular problem and set it up properly. In addition, there is no optimal algorithm to solve all prediction problems. Thus, in order to improve the result of the classification process, some techniques combine the knowledge acquired by individual learning algorithms aiming to discover new patterns not yet identified. Among these techniques, there is the stacking strategy. This strategy consists in the combination of outputs of base classifiers, induced by several learning algorithms using the same dataset, by means of another classifier called meta-classifier. This paper aims to verify the relation between the classifiers diversity and the quality of stacking. We have performed a lot of experiments which results show the impact of multiple diversity measures on the gain of stacking.

A Study on the Relationship between Internal and External Validity Indices Applied to Partitioning and Density-based Clustering Algorithms.

Abstract: Measuring the quality of data partitions is essential to the success of clustering applications. A lot of different validity indices have been proposed in the literature, but choosing the appropriate index for evaluating the results of a particular clustering algorithm remains a challenge. Clustering results can be evaluated using different indices based on external or internal criteria. An external criterion requires a partitioning of the data previously defined for comparison with the clustering results while an internal criterion evaluates clustering results considering only the data properties. In a previous work we proposed a method for selecting the most suitable cluster validity internal index applied on the results of partitioning clustering algorithms. In this paper we extend our previous work validating the method for density-based clustering algorithms. We have looked into the relationships between internal and external indices, relating them through linear regression and regression model trees. Each algorithm was run over synthetic datasets generated for this purpose, using different configurations. Experiments results point out that Silhouette and Gamma are the most suitable indices for evaluating both the datasets with compactness property and the datasets with multiple density.

A Fuzzy Approach to Measure the Similarity Between Web Streaming Users

The increase of the Internet access and the popularity of mobile devices have influenced the consumption of radio/TV programs on the Web. An alternative for customization programming is the use of recommendation systems to adapt the content transmitted based on the preference of the listeners. The behavior of users accessing content on the Web is highly uncertain and naturally diffuse. In this paper, we propose an approach based on fuzzy set theory to analyze the similarity between users of Web radio programs, capturing similar interests from streaming data available in log files.