Celso A. A. Kaestner

Automatic Text Summarization Using a Machine Learning Approach

In this paper we address the automatic summarization task. Recent research works on extractive-summary generation employ some heuristics, but few works indicate how to select the relevant features. We will present a summarization procedure based on the application of trainable Machine Learning algorithms which employs a set of features extracted directly from the original text. These features are of two kinds: statistical - based on the frequency of some elements in the text; and linguistic - extracted from a simplified argumentative structure of the text. We also present some computational results obtained with the application of our summarizer to some well known text databases, and we compare these results to some baseline summarization procedures.

Attribute Selection with a Multi-objective Genetic Algorithm

In this paper we address the problem of multi-objective attribute selection in data mining. We propose a multi-objective genetic algorithm (GA) based on the wrapper approach to discover the best subset of attributes for a given classification algorithm, namely C4.5, a well-known decision-tree algorithm. The two objectives to be minimized are the error rate and the size of the tree produced by C4.5. The proposed GA is a multi-objective method in the sense that it discovers a set of non-dominated solutions (attribute subsets), according to the concept of Pareto dominance.

On the suitability of state-of-the-art music information retrieval methods for analyzing, categorizing and accessing non-Western and ethnic music collections

With increasing amounts of music being available in digital form, research in music information retrieval has turned into a dominant field to support organization of and easy access to large collections of music. Yet, most research is focussed traditionally on Western music, mostly in the form of mastered studio recordings. This leaves the question whether current music information retrieval approaches can also be applied to collections of non-Western and in particular ethnic music with completely different characteristics and requirements. In this work we analyze the performance of a range of automatic audio description algorithms on three music databases with distinct characteristics, specifically a Western music collection used previously in research benchmarks, a collection of Latin American music with roots in Latin American culture, but following Western tonality principles, as well as a collection of field recordings of ethnic African music. The study quantitatively shows the advantages and shortcomings of different feature representations extracted from music on the basis of classification tasks, and presents an approach to visualize, access and interact with ethnic music collections in a structured way.

Generating Text Summaries through the Relative Importance of Topics

This work proposes a new extractive text-summarization algorithm based on the importance of the topics contained in a document. The basic ideas of the proposed algorithm are as follows. At first the document is partitioned by using the TextTiling algorithm, which identifies topics (coherent segments of text) based on the TF-IDF metric. Then for each topic the algorithm computes a measure of its relative relevance in the document. This measure is computed by using the notion of TF-ISF (Term Frequency - Inverse Sentence Frequency), which is our adaptation of the well-known TF-IDF (Term Frequency - Inverse Document Frequency) measure in information retrieval. Finally, the summary is generated by selecting from each topic a number of sentences proportional to the importance of that topic.

Automatic music genre classification using ensemble of classifiers

This paper presents a novel approach to the task of automatic music genre classification which is based on multiple feature vectors and ensemble of classifiers. Multiple feature vectors are extracted from a single music piece. First, three 30-second music segments, one from the beginning, one from the middle and one from end part of a music piece are selected and feature vectors are extracted from each segment. Individual classifiers are trained to account for each feature vector extracted from each music segment. At the classification, the outputs provided by each individual classifier are combined through simple combination rules such as majority vote, max, sum and product rules, with the aim of improving music genre classification accuracy. Experiments carried out on a large dataset containing more than 3,000 music samples from ten different Latin music genres have shown that for the task of automatic music genre classification, the features extracted from the middle part of the music provide better results than using the segments from the beginning or end part of the music. Furthermore, the proposed ensemble approach, which combines the multiple feature vectors, provides better accuracy than using single classifiers and any individual music segment.

Feature Selection in Automatic Music Genre Classification

This paper presents the results of the application of a feature selection procedure to an automatic music genre classification system. The classification system is based on the use of multiple feature vectors and an ensemble approach, according to time and space decomposition strategies. Feature vectors are extracted from music segments from the beginning, middle and end of the original music signal (time decomposition). Despite being music genre classification a multi-class problem, we accomplish the task using a combination of binary classifiers, whose results are merged in order to produce the final music genre label (space decomposition). As individual classifiers several machine learning algorithms were employed: naive-Bayes, decision trees, support vector machines and multi-layer perceptron neural nets. Experiments were carried out on a novel dataset called Latin music database, which contains 3,227 music pieces categorized in 10 musical genres. The experimental results show that the employed features have different importance according to the part of the music signal from where the feature vectors were extracted. Furthermore, the ensemble approach provides better results than the individual segments in most cases.

Automatic Bird Species Identification for Large Number of Species

In this paper we focus on the automatic identification of bird species from their audio recorded song. Bird monitoring is important to perform several tasks, such as to evaluate the quality of their living environment or to monitor dangerous situations to planes caused by birds near airports. We deal with the bird species identification problem using signal processing and machine learning techniques. First, features are extracted from the bird recorded songs using specific audio treatment, next the problem is performed according to a classical machine learning scenario, where a labeled database of previously known bird songs are employed to create a decision procedure that is used to predict the species of a new bird song. Experiments are conducted in a dataset of recorded songs of bird species which appear in a specific region. The experimental results compare the performance obtained in different situations, encompassing the complete audio signals, as recorded in the field, and short audio segments (pulses) obtained from the signals by a split procedure. The influence of the number of classes (bird species) in the identification accuracy is also evaluated.

Automatic Text Summarization with Genetic Algorithm-Based Attribute Selection

The task of automatic text summarization consists of generating a summary of the original text that allows the user to obtain the main pieces of information available in that text, but with a much shorter reading time. This is an increasingly important task in the current era of information overload, given the huge amount of text available in documents. In this paper the automatic text summarization is cast as a classification (supervised learning) problem, so that machine learning-oriented classification methods are used to produce summaries for documents based on a set of attributes describing those documents. The goal of the paper is to investigate the effectiveness of Genetic Algorithm (GA)-based attribute selection in improving the performance of classification algorithms solving the automatic text summarization task. Computational results are reported for experiments with a document base formed by news extracted from The Wall Street Journal of the TIPSTER collection–a collection that is often used as a benchmark in the text summarization literature.

Generation of Signatures by Deformations

The techniques of automatic classification of hand-written signatures have been studied and some of them are based on the application of neuronal nets or statistical methods. Nevertheless, the great number of samples required by these methods turns many of its practical applications unfeasible. This article describes a technique for automatic generation of signatures originated from the deformation of a reduced number of genuine samples. The technique used here is based on convolution between deforming polynomials representing the deformations and the signals representing the horizontal and vertical moves of the pen, required for the reproduction of the original samples. The result of the convolution produces the deformation of those signals and, consequently, the deformation of the tracing obtained from them.

An optimized hill climbing algorithm for feature subset selection: evaluation on handwritten character recognition

This paper presents an optimized Hill-Climbing algorithm to select subset of features for handwritten character recognition. The search is conducted taking into account a random mutation strategy and the initial relevance, of each feature in the recognition process. A first set of experiments have shown a reduction in the original number of features used in an MLP-based character recognizer from 132 to 77 features (reduction of 42%) without a significant loss in terms of recognition rates, which are 99.1% for 30,089 digits and 93.0% for 11,941 uppercase characters, both handwritten samples from the NIST SD19 database. Additional experiments have been done by considering some loss in terms of recognition rate during the feature subset selection. A byproduct of these experiments is a cascade classifier based on feature subsets of different sizes, which is used to reduce the complexity of the classification task by 86.54% on the digit recognition experiment. The proposed feature selection method has shown to be an interesting strategy to implement a wrapper approach without the need of complex and expensive hardware architectures.