International Journal on Artificial Intelligence Tools

Effective feature selection methods are important for improving the efficiency and accuracy of text categorization algorithms by removing redundant and irrelevant terms from the corpus. Extensive research has been done to improve the performance of individual feature selection methods. However, it is always a challenge to come up with an individual feature selection method which would outperform other methods in most cases. In this paper, we explore the possibility of improving the overall performance by combining multiple individual feature selection methods. In particular, we propose a method of combining multiple feature selection methods by using an information fusion paradigm, called Combinatorial Fusion Analysis (CFA). A rank-score function and its associated graph, called rank-score graph, are adopted to measure the diversity of different feature selection methods. Our experimental results demonstrated that a combination of multiple feature selection methods can outperform a single method only if each individual feature selection method has unique scoring behavior and relatively high performance. Moreover, it is shown that the rank-score function and rank-score graph are useful for the selection of a combination of feature selection methods.

Virtual Organizations are a mechanism where agents can demonstrate their social skills since they can work in a cooperative and collaborative way. Nonetheless, the development of organizations using Multi-Agent Systems (MAS) requires extensive experience in different methodologies and platforms. Model-Driven Development (MDD) is a technique for generating application code that is developed from basic models and meta-models using a variety of automatic transformations. This paper presents an approach to develop and deploy organization-oriented Multi-Agent Systems using a model-driven approach. Based on this idea, we introduce a relatively generic agent-based meta-model for a Virtual Organization, which was created by a comprehensive analysis of the organization-oriented methodologies used in MAS. Following the MDD approach, the concepts and relationships obtained were mapped into two different platforms available for MAS development, allowing the validation of our proposal. In this way, the resultant approach can generate Virtual Organization deployments from unified meta-models, facilitating the development process of agent-based software from the user point of view.

In this work, we propose a new measure of semantic relatedness between concepts applied in word sense disambiguation. Using the overlaps between WordNet definitions of concepts (glosses) and the so-called goodness of fit statistical test we establish a formal mechanism for quantifying and estimating the semantic relatedness between concepts. More concretely, we model WordNet glosses overlaps by making a theoretical assumption about their distribution and then we quantify the discrepancy between the theoretical and actual distribution. This discrepancy is suitably used to measure the relatedness between the input concepts. The experimental results showed very good performance on SensEval-2 lexical sample data for word sense disambiguation.

The recent years have been marked by a rapid growth in the World Wide Web 2.0 applications such as blog posts, forums, mailing lists, and product-review websites. As a result, a special sentiment analysis field has sprung up relevant to the issue of people's responses to the diversity of available subjects. Hence, one might well wonder: how do people feel and react when dealing with certain topics? In this paper, a new automatic sentiment-processing model has been advanced, whereby the current problems faced by the prevalent existing models can be deciphered and more properly treated. The suggested approach consists in developing a multi-agent system based on a thorough linguistic analysis, meanwhile highlighting the major contributions provided by such a study in combination with the syntactic, semantic, and subjective analyses. Actually, the newly-devised framework enables to resolve the ambiguities and complexities of the natural evaluative language and to strengthen, as well as consolidate, the results achieved at the various analysis stages thereof.

Due to the infeasibility of randomized controlled experiments, the existence of unobserved variables and the fact that equivalent direct acyclic graphs obtained generally can not be distinguished, it is difficult to learn the true causal relations of original graph. This paper presents an algorithm called BSPC based on adjacent nodes to learn the structure of Causal Bayesian Networks with unobserved variables by using observational data. It does not have to adjust the structure as the existing algorithms FCI and MBCS*, while it can guarantee to obtain the true adjacent nodes. More important is that algorithm BSPC reduces computational complexity and improves reliability of conditional independence tests. Theoretical results show that the new algorithm is correct. In addition, the advantages of BSPC in terms of the number of conditional independence tests and the number of orientation errors are illustrated with simulation experiments from which we can see that it is more suitable in order to learn the structure of Causal Bayesian Networks with latent variables. Moreover a better latent structure representation is returned.

Many combinatorial problems can be formulated as Valued Constraint Satisfaction Problems (VCSPs). In this framework, the constraints are defined by means of valuation functions to reflect several degrees of coherence. Despite the NP-hardness of the VCSP, tractable versions can be obtained by forcing the allowable valuation functions to have specific features. This is the case for submodular VCSPs, i.e. VCSPs that involve submodular valuation functions only.

In this paper, we propose a problem decomposition scheme for binary VCSPs that takes advantage of submodular functions even when the studied problem is not submodular. The proposed scheme consists in decomposing the problem to be solved into a set of submodular, then tractable, subproblems. The decomposition scheme combines two techniques that where already used in the framework of constraint-based reasoning, but in separate manner. These techniques are domain partitioning and value permutation.

Paraphrase Recognition systems most often use various lexical, syntactic and semantic features to recognize paraphrases. This paper presents the work done in designing a Support Vector Machine (SVM) based Paraphrase Recognizer and then improving its performance using feature selection strategy. Wrapper method of feature selection has been adopted by combining Genetic Algorithms with Support Vector Machine Classifiers. Experimental results show that applying Feature selection improves the accuracy besides reducing the number of features. The developed paraphrase recognizer has been applied for the Student Answer Evaluation task. The results obtained show that the performance of Answer Evaluation systems which use only half the number of features is comparable to systems using the original feature set.

Imbalanced data sets in the class distribution is common to many real world applications. As many classifiers tend to degrade their performance over the minority class, several approaches have been proposed to deal with this problem. In this paper, we propose two new cluster-based oversampling methods, SOI-C and SOI-CJ. The proposed methods create clusters from the minority class instances and generate synthetic instances inside those clusters. In contrast with other oversampling methods, the proposed approaches avoid creating new instances in majority class regions. They are more robust to noisy examples (the number of new instances generated per cluster is proportional to the cluster's size). The clusters are automatically generated. Our new methods do not need tuning parameters, and they can deal both with numerical and nominal attributes. The two methods were tested with twenty artificial datasets and twenty three datasets from the UCI Machine Learning repository. For our experiments, we used six classifiers and results were evaluated with recall, precision, F-measure, and AUC measures, which are more suitable for class imbalanced datasets. We performed ANOVA and paired t-tests to show that the proposed methods are competitive and in many cases significantly better than the rest of the oversampling methods used during the comparison.

It has been demonstrated that human users attribute a personality to the computer interfaces they use, regardless of whether one has been explicitly encoded in the system's design or not. In this paper, we explore a method for having explicit control over the personality that a spoken human-robot interface is perceived to exhibit by its users. Our method focuses on the interaction between users and semantic knowledge-based systems where the goal of the interaction is that information from the semantic store is relayed to the user. We describe a personality modelling method that complements a standard dialogue manager by calculating parameters related to adaptivity and emotion for the various interaction modules that realize the system's dialogue acts. This calculation involves the planned act, the user adaptivity model, the system's own goals, but also a machine representation of the personality that we want the system to exhibit, so that systems with different personality will react differently even when in the same dialogue state and with the same user or user type.