Leandro Nunes de Castro

A self-organizing neural network using ideas from the immune system to solve the traveling salesman problem

Most combinatorial optimization problems belong to the NP-complete or NP-hard classes, which means that they may require an infeasible processing time to be solved by an exhaustive search method. Thus, less expensive heuristics in respect to the processing time are commonly used. These heuristics can obtain satisfactory solutions in short running times, but there is no guarantee that the optimal solution will be found. Artificial Neural Networks (ANNs) have been widely studied to solve combinatorial problems, presenting encouraging results. This paper proposes some modifications on RABNET-TSP, an immune-inspired self-organizing neural network, for the solution of the Traveling Salesman Problem (TSP). The modified algorithm is compared with other neural methods from the literature and the results obtained suggest that the proposed method is competitive in relation to the other ones, outperforming them in many cases with regards to the quality (cost) of the solutions found, though demanding a greater time for convergence in many cases.

A keyword extraction method from twitter messages represented as graphs

Abstract Twitter is a microblog service that generates a huge amount of textual content daily. All this content needs to be explored by means of text mining, natural language processing, information retrieval, and other techniques. In this context, automatic keyword extraction is a task of great usefulness. A fundamental step in text mining techniques consists of building a model for text representation. The model known as vector space model, VSM, is the most well-known and used among these techniques. However, some difficulties and limitations of VSM, such as scalability and sparsity, motivate the proposal of alternative approaches. This paper proposes a keyword extraction method for tweet collections that represents texts as graphs and applies centrality measures for finding the relevant vertices (keywords). To assess the performance of the proposed approach, three different sets of experiments are performed. The first experiment applies TKG to a text from the Time magazine and compares its performance with that of the literature. The second set of experiments takes tweets from three different TV shows, applies TKG and compares it with TFIDF and KEA, having human classifications as benchmarks. Finally, these three algorithms are applied to tweets sets of increasing size and their computational running time is measured and compared. Altogether, these experiments provide a general overview of how TKG can be used in practice, its performance when compared with other standard approaches, and how it scales to larger data instances. The results show that TKG is a novel and robust proposal to extract keywords from texts, particularly from short messages, such as tweets.

Clustering algorithm selection by meta-learning systems

Data clustering aims to segment a database into groups of objects based on the similarity among these objects. Due to its unsupervised nature, the search for a good-quality solution can become a complex process. There is currently a wide range of clustering algorithms, and selecting the best one for a given problem can be a slow and costly process. In 1976, Rice formulated the Algorithm Selection Problem (ASP), which postulates that the algorithm performance can be predicted based on the structural characteristics of the problem. Meta-learning brings the concept of learning about learning; that is, the meta-knowledge obtained from the algorithm learning process allows the improvement of the algorithm performance. Meta-learning has a major intersection with data mining in classification problems, in which it is normally used to recommend algorithms. The present paper proposes new ways to obtain meta-knowledge for clustering tasks. Specifically, two contributions are explored here: (1) a new approach to characterize clustering problems based on the similarity among objects; and (2) new methods to combine internal indices for ranking algorithms based on their performance on the problems. Experiments were conducted to evaluate the recommendation quality. The results show that the new meta-knowledge provides high-quality algorithm selection for clustering tasks.

Bio-inspired and gradient-based algorithms to train MLPs: The influence of diversity

This paper has three main goals: (i) to employ two classes of algorithms: bio-inspired and gradient-based to train multi-layer perceptron (MLP) neural networks for pattern classification; (ii) to combine the trained neural networks into ensembles of classifiers; and (iii) to investigate the influence of diversity in the classification performance of individual and ensembles of classifiers. The optimization version of an artificial immune network, named opt-aiNet, particle swarm optimization (PSO) and an evolutionary algorithm (EA) are used as bio-inspired methods to train MLP networks. Besides, the standard backpropagation with momentum (BPM), a quasi-Newton method called DFP and a modified scaled-conjugate gradient (SCGM) are the gradient-based algorithms used to train MLP networks in this work. Comparisons among all the training methods are presented in terms of classification accuracy and diversity of the solutions found. The results obtained suggest that most bio-inspired algorithms deteriorate the diversity of solutions during the search, while immune-based methods, like opt-aiNet, and multiple initializations of standard gradient-based algorithms provide diverse solutions that result in good classification accuracy for the ensembles.

A polarity analysis framework for Twitter messages

Social media, such as Twitter and Facebook, allow the creation, sharing and exchange of information among people, companies and brands. This information can be used for several purposes, such as to understand consumers and their preferences. In this direction, the sentiment analysis can be used as a feedback mechanism. This analysis corresponds to classifying a text according to the sentiment that the writer intended to transmit. A basic sentiment classifier determines the sentiment polarity (negative, neutral or positive) of a given text at the document, sentence, or feature/aspect level. Advanced types may consider other elements like the emotional state (e.g. angry, sad, happy), affective states (e.g. pleasure and pain), motivational states (e.g. hunger and curiosity), temperaments, among others. In general, there are two main approaches to attribute sentiment to tweets: based on knowledge; or based on machine learning algorithms. In the latter case, the learning algorithm requires a pre-classified data sample to determine the class of new data. Typically, the sample is pre-classified manually, making the process time consuming and reducing its real time applicability for big data. This paper proposes a polarity analysis framework for Twitter messages, which combines both approaches and an automatic contextual module. To assess the performance of the proposed framework, four text datasets from the literature are used. Five different types of classifiers were considered: Naive Bayes (NB); Support Vector Machines (SVM); Decision Trees (J48); and Nearest Neighbors (KNN). The results show that the proposal is a suitable framework to automate the whole polarity analysis process, providing high accuracy levels and low false positive rates.

Automatic sentiment analysis of Twitter messages

Twitter® is a microblogging service usually used as an instant communication platform. The capacity to provide information in real time has stimulated many companies to use this service to understand their consumers. In this direction, TV stations have adopted Twitter for shortening the distance between them and their viewers, and use such information as a feedback mechanism for their shows. The sentiment analysis task can be used as one such feedback mechanism. This task corresponds to classifying a text according to the sentiment that the writer intended to transmit. A classifier usually requires a pre-classifled data sample to determine the class of new data. Typically, the sample is pre-classified manually, making the process time consuming and reducing its real time applicability for big data. This paper proposes an automatic sentiment classifier for Twitter messages, and uses TV shows from Brazilian stations for benchmarking. The automatic sentiment analysis reduces human intervention and, thus, the complexity and cost of the whole process. To assess the performance of the proposed system tweets related to a Brazilian TV show were captured in a 24h interval and fed into the system. The proposed technique achieved an average accuracy of 90%.

BeeRBF: A bee-inspired data clustering approach to design RBF neural network classifiers

Different methods have been used to train radial basis function neural networks. This paper proposes a bee-inspired algorithm to automatically select the number and location of basis functions to be used in such RBF network. The algorithm was designed to solve data clustering problems, where the centroids of clusters are used as centers for the RBF network. The approach presented in this paper is preliminary evaluated in three synthetic datasets, two classification datasets and one function approximation problem, and its results suggest a potential for real-world application.

Bee colonies as model for multimodal continuous optimization: The OptBees algorithm

This paper presents the OptBees, an optimization algorithm inspired by the processes of collective decision-making by bee colonies. The algorithm was designed with the objective of generating and maintaining diversity, promoting a multimodal search and obtaining multiple local optima without losing the ability of global optimization, thus representing an innovation compared with existent bee-inspired algorithms. It has been tested in five of the twenty-five minimization problems proposed for the Optimization Competition of Real Parameters of the CEC 2005 Special Session on Real-Parameter Optimization, held in the 2005 IEEE Congress on Evolutionary Computation (CEC). The results obtained suggest the suitability of the algorithm to exploit multimodality of problems, being successful in the generation and maintenance of diversity and achieving good quality results in global optimization.

The proposal of a fuzzy clustering algorithm based on particle swarm

This paper proposes the Fuzzy Particle Swarm Clustering (FPSC) algorithm, which is an extension of the crisp data clustering algorithm PSC particularly tailored to deal with fuzzy clusters. The main structural changes of the original PSC algorithm to design FPSC occurred in the selection and evaluation steps of the winner particle, comparing the degree of membership of each object from the database in relation to the particles in the swarm. The FPSC algorithm was applied to eight databases from the literature with the purpose of benchmarking and its performance was compared with that of Fuzzy C-Means and Fuzzy PSO. The results showed that the FPSC algorithm is competitive with the algorithms discussed in this paper.

Bladder Carcinoma Data with Clinical Risk Factors and Molecular Markers: A Cluster Analysis

Bladder cancer occurs in the epithelial lining of the urinary bladder and is amongst the most common types of cancer in humans, killing thousands of people a year. This paper is based on the hypothesis that the use of clinical and histopathological data together with information about the concentration of various molecular markers in patients is useful for the prediction of outcomes and the design of treatments of nonmuscle invasive bladder carcinoma (NMIBC). A population of 45 patients with a new diagnosis of NMIBC was selected. Patients with benign prostatic hyperplasia (BPH), muscle invasive bladder carcinoma (MIBC), carcinoma in situ (CIS), and NMIBC recurrent tumors were not included due to their different clinical behavior. Clinical history was obtained by means of anamnesis and physical examination, and preoperative imaging and urine cytology were carried out for all patients. Then, patients underwent conventional transurethral resection (TURBT) and some proteomic analyses quantified the biomarkers (p53, neu, and EGFR). A postoperative follow-up was performed to detect relapse and progression. Clusterings were performed to find groups with clinical, molecular markers, histopathological prognostic factors, and statistics about recurrence, progression, and overall survival of patients with NMIBC. Four groups were found according to tumor sizes, risk of relapse or progression, and biological behavior. Outlier patients were also detected and categorized according to their clinical characters and biological behavior.