No abstract

International audienceNaive Bayesian Classifiers, which rely on independence hypotheses, together with a normality assumption to estimate densities for numerical data, are known for their simplicity and their effectiveness. However, estimating densities, even under the normality assumption, may be problematic in case of poor data. In such a situation, possibility distributions may provide a more faithful representation of these data. Naive Possibilistic Classifiers (NPC), based on possibility theory, have been recently proposed as a counterpart of Bayesian classifiers to deal with classification tasks. There are only few works that treat possibilistic classification and most of existing NPC deal only with categorical attributes. This work focuses on the estimation of possibility distributions for continuous data. In this paper we investigate two kinds of possibilistic classifiers. The first one is derived from classical or flexible Bayesian classifiers by applying a probability–possibility transformation to Gaussian distributions, which introduces some further tolerance in the description of classes. The second one is based on a direct interpretation of data in possibilistic formats that exploit an idea of proximity between data values in different ways, which provides a less constrained representation of them. We show that possibilistic classifiers have a better capability to detect new instances for which the classification is ambiguous than Bayesian classifiers, where probabilities may be poorly estimated and illusorily precise. Moreover, we propose, in this case, an hybrid possibilistic classification approach based on a nearest-neighbour heuristics to improve the accuracy of the proposed possibilistic classifiers when the available information is insufficient to choose between classes. Possibilistic classifiers are compared with classical or flexible Bayesian classifiers on a collection of benchmarks databases. The experiments reported show the interest of possibilistic classifiers. In particular, flexible possibilistic classifiers perform well for data agreeing with the normality assumption, while proximity-based possibilistic classifiers outperform others in the other cases. The hybrid possibilistic classification exhibits a good ability for improving accuracy

Automatic text summarization is the process of generating or extracting a brief representation of an input text. There are several algorithms for extractive summarization in the literature tested by using English and other languages datasets; however, only few extractive Arabic summarizers exist due to the lack of large collection in Arabic language. This paper proposes and assesses new extractive single-document summarization approaches based on analogical proportions which are statements of the form "a is to b as c is to d". The goal is to study the capability of analogical proportions to represent the relationship between documents and their corresponding summaries. For this purpose, we suggest two algorithms to quantify the relevance/irrelevance of an extracted keyword from the input text, to build its summary. In the first algorithm, the analogical proportion representing this relationship is limited to check the existence/non-existence of the keyword in any document or summary in a binary way without considering keyword frequency in the text, whereas the analogical proportion of the second algorithm considers this frequency. We have assessed and compared these two algorithms with some languageindependent summarizers (LexRank, TextRank, Luhn and LSA (Latent Semantic Analysis)) using our large corpus ANT (Arabic News Texts) and a small test collection EASC (Essex Arabic Summaries Corpus) by computing ROUGE (Recall-Oriented Understudy for Gisting Evaluation) and BLEU (BiLingual Evaluation Understudy) metrics. The best-achieved results are ROUGE-1 = 0.96 and BLEU-1 = 0.65 corresponding to educational documents from EASC collection which outperform the best LexRank algorithm. The proposed algorithms are also compared with three other Arabic extractive summarizers, using EASC collection, and show better results in terms of ROUGE-1 = 0.75 and BLEU-1 = 0.47 for the first algorithm, and ROUGE-1 = 0.74 and BLEU-1 = 0.49 for the second one. Experimental results show the interest of analogical proportions for text summarization. In particular, analogical summarizers significantly outperform three among four language-independent summarizers in the case of BLEU-1 for ANT collection and they are not significantly outperformed by any other summarizer in the case of EASC collection.

No abstract

In real-world problems, input data may be pervaded with uncertainty. In this paper, we investigate the behavior of naive possibilistic classifiers, as a counterpart to naive Bayesian ones, for dealing with classification tasks in presence of uncertainty. For this purpose, we extend possibilistic classifiers, which have been recently adapted to numerical data, in order to cope with uncertainty in data representation. Here the possibility distributions that are used are supposed to encode the family of Gaussian probabilistic distributions that are compatible with the considered data set. We consider two types of uncertainty: i) the uncertainty associated with the class in the training set, which is modeled by a possibility distribution over class labels, and ii) the imprecision pervading attribute values in the testing set represented under the form of intervals for continuous data. Moreover, the approach takes into account the uncertainty about the estimation of the Gaussian distribution parameters due to the limited amount of data available. We first adapt the possibilistic classification model, previously proposed for the certain case, in order to accommodate the uncertainty about class labels. Then, we propose an algorithm based on the extension principle to deal with imprecise attribute values. The experiments reported show the interest of possibilistic classifiers for handling uncertainty in data. In particular, the probability-to-possibility transform-based classifier shows a robust behavior when dealing with imperfect data.