Abdellah Madani

Building a syntactic rules-based stemmer to improve search effectiveness for arabic language

Nowadays, The world is experiencing a huge growth in the volume of exchanged texts, which makes some of it untapped. Text Mining is the set of techniques that analyze these large masses of information, extract relations that can be unknown beforehand and provide solutions that help decision making. In this sense, stemming is a common requirement of these techniques. It includes reducing different grammatical forms of a word and bringing them to a common base form. In what follows, we will discuss these treatment methods for arabic text, show their limits and provide new algorithm to improve them.

A new modeling approach for Arabic opinion mining recognition

Over recent years, the world has experienced a huge growth in the volume of shared web texts. Its users generate daily a huge volume of comments and reviews related to different aspects of their lives. In general, opinion mining/sentiment analysis refers to the task of identifying positive and negative opinions, emotions and evaluations related to an article, news, products, services, etc [1]. Arabic Opinion mining is conducted in this study using a dataset consisting of 625 Arabic reviews and comments collected from Trip Advisor website. We introduce a new mathematical approach to recognize authors opinion. As the weights computation is determining in the classification, we formulate first a linear program to maximize the distance between the considered classes, then we use these weights to calculate the label of each comment. A further post optimization is also treated to add other contributing descriptors in order to adjust the classification. The results which based on Support Vector Machines showed that the approach is the most influencing on opinion recognition.

Integrating effective rules to improve arabic text stemming

Nowadays, with the growth in the use of search engines, the extension of spying programs and anti -terrorism prevention, several researches focused on text analysis. In this sense, lemmatization and stemming are two common requirements of these researches. They include reducing different grammatical forms of a word and bring them to a common base form. In what follows, we will discuss these treatment methods on arabic text, especially the Khoja Stemmer, show their limits and provide new tools to improve it.

New rules-based algorithm to improve Arabic stemming accuracy

In the recent past, the world has been witnessing a steady increase in the area of natural language processing owing to the spread of the internet. However, attempts and efforts devoted for Arabic language are still limited. By morphological and semantic properties, Arabic is considered a difficult language in the field of automatic processing. From that perspective, many different approaches were attempted to deal with the morphological variation and the agglutination phenomenon while stemming Arabic texts. Formally, stemming and light-stemming are used to remove irrelevant morphological variations from a given word, and extract its original stem or root. This research introduces a complete new rules-based algorithm. This involves precise removal of affixes based on context-sensitive morphological rules and then deduces the root according to a predefined set of rules. Finally, results show that the accuracy of the proposed algorithm is higher than the two well-known Arabic stemmers.

An improved Chi-sqaure feature selection for Arabic text classification using decision tree

Feature selection is an important and necessary step that can improve greatly the classification performance. The aim of the present paper is to investigate a new feature selection (referred to, hereafter, as ImpCHI), when using light stemming. ImpCHI is an improvement of chi-square - one of the most effective feature selection methods to date. Evaluation used a corpus that consists of 250 Arabic documents independently classified into five classes: art and culture, economics, politics, society, and sport. The experiment results show that Arabic text classification using ImpCHI as feature selection outperforms using chi-square in terms of recall-measures.

A hybrid optimal weighting scheme and machine learning for rendering sentiments in tweets

Over recent years, the world has experienced an explosive growth in the volume of shared web texts. Everyday, a huge volume of opinions expressed in various forms such as articles, reviews and tweets is generated. In general, opinion mining refers to the task of extracting opinions, and sentiment analysis is the technique that extracts subjectivity and polarity; in other words, it determines whether a text is positive or negative Taboada et al., 2011. Arabic sentiment analysis is conducted in this study using a publically available data set written in both modern standard Arabic and the Jordanian dialect. A new mathematical approach is introduced to determine the polarity of the tweet by using four functions whose parameters are the solutions of a linear program. These functions are then classified using support vector machines and K-nearest neighbours. The results show that the proposed approach is considerably reliable in Arabic sentiment analysis.

A combination of low-level light stemming and support vector machines for the classification of Arabic opinions

Recent years have brought the burst of volume of shared opinionated texts across the internet. Every day, a tremendous number of comments and reviews towards different aspects of our lives is generated through social networks and other websites. A large portion of these data is written in Arabic which is the fifth most used language on internet and is one of the six official languages of the United Nations. However, the templatic morphology of Arabic language makes it a difficult language in the fields of information retrieval. In light of the scarcity of Arabic opinion mining systems, this paper introduces a complete approach to prepare, model and classify Arabic comments and reviews. It is conducted using a dataset consisting of 500 comments collected from TripAdvisor Website which fall into four scaled classes of opinions. For this, a new stemming technique has been defined. Stemmed opinions are then quantified according to a set of mathematical functions, and classified by different machine learning techniques. Support vector machines have yielded the highest f-measure. Finally, the impact of each step on the overall performance has been evaluated.

New distances combination for facial expression recognition from image sequences

This paper propose a new method of facial expression recognition from image sequences taking into consideration only the first and the last frames, that represent respectively the neutral state and the emotion state. Our proposed method is based on calculating six distances from eleven points of the face. These points are detected and tracked using supervised decent method, and refer to the four internal parts of the face (eyebrows, eyes, nose, and mouth). After that, two data mining techniques are used to classify facial expression into emotion, taking as input dynamic feature vector deducted from the six distances previously calculated in the first and the last frames. Our method achieved a best recognition rate of 98% in the CK+ database, and 80.6% in the OULU-CASIA VIS database.

Optimizing self-organizing method in traffic lights with PSO algorithm in a single junction modeled with a cellular automaton

Among raised solutions for global flow optimization in traffic jam, we can quote traffic lights optimization. Several researches treated this problem and proposed different models to solve it. In this paper, we will use the cellular automaton to model the traffic flow in crossroads and the self-organizing method to control traffic lights and then we will apply the PSO algorithm to get the appropriate values of parameters for self-organizing method. We will demonstrate that PSO algorithm provides an optimal solution in order to improve the global flow.