• Digital Humanities and Cultural Heritage
• Publications close
• Conference object close
• KW close

Inevitably generating a robust summary from a long Arabic document is a challenging task owing to the fact that Arabic is a complex language and has unique attributes. In this paper, we propose an integrated approach between abstractive and extractive for providing an informative and coherent summary from a long document. The extractive method employs a novel formulation for extracting a set of statistical and semantic features by taking into consideration the semantic, importance, and position of the sentence. The combination of statistical and semantic features is used to learn a soft voting classifier to extract the significant sentences. In the abstractive approach, only significant sentences that classified from the extractive approach will be trained with encoder-decoder bidirectional long short-term memory (LSTM) for producing a compose novel summary. We show that the mixed proposed architecture between extractive and abstractive outperforms and provides better results comparing to some existing Arabic summarizing systems.

This paper describes SalamNET, an Arabic offensive language detection system that has been submitted to SemEval 2020 shared task 12: Multilingual Offensive Language Identification in Social Media. Our approach focuses on applying multiple deep learning models and conducting in depth error analysis of results to provide system implications for future development considerations. To pursue our goal, a Recurrent Neural Network (RNN), a Gated Recurrent Unit (GRU), and Long-Short Term Memory (LSTM) models with different design architectures have been developed and evaluated. The SalamNET, a Bi-directional Gated Recurrent Unit (Bi-GRU) based model, reports a macro-F1 score of 0.83. Comment: In Proceedings of the International Workshop on Semantic Evaluation (SemEval) 2020

The Vector Space Model (VSM) is a common document representation model that is widely used in data mining and information retrieval (IR) systems. However, this technique poses some challenges such as high dimensional space and semantic loss representation. Therefore, the latent semantic indexing (LSI) is proposed to reduce the feature dimensions and to generate semantic rich features that represent conceptual term-document associations. In particular, LSI has been successfully implemented in search engines and text classification tasks. In this paper, we propose a novel approach to enhance the quality of the retrieved documents in search engines for Arabic language. That is, we propose to use a new extension of the LSI technique instead of just using the standard LSI technique. The LSI method proposed is based on employing the word co-occurrences to form a term-by-document matrix. The proposed method is to be based on the documents evaluating cosine similarity measures for term-by-document matrix. We will empirically evaluate the performance using an Arabic data collection that contains no less than 500 documents with no less than 30,000 unique words. A testing set contains keywords from a specific domain will be used to evaluate the quality of the top 20-30 retrieved documents using different singular values (i.e. different number of dimensions). The results will be judged on the performance of the proposed method as it is compared to the standard LSI.

Based on the data analysis of ultrasonic experiments, a novel approach has been developed to explore Jahn-Teller effect (JTE) problems in non-cubic crystals with JT centers without involving additional experimental data beyond the information about the electronic term and crystal symmetry. Distinguished from cubic crystals, the axis of symmetry of the bulk non-cubic crystal do not necessarily coincide with those of the local impurity center, thus complicating the relation between the distortions produced by the ultrasound wave and the JTE active modes. We analysed the problem with corresponding calculations for the wurtzite-type hexagonal crystal CdSe:Cr 2+ , in which the chromium ion substitutes the cadmium one in the tetrahedral environment, resulting in its electronic ground state 5 T 2 (e 2 t 2 ). Experimental investigation of this system by ultrasound at frequencies of 28-105 MHz in the temperature range of 4-180 K, yields a peak in the attenuation of the ultrasound below 40 K for the normal modes related to the c 11 , c 44 , c 55 , c 55 , and c 66 elastic moduli. The peak has been interpreted as the manifestation of the JTE, similar to the one, observed in cubic crystals doped with 3d ions. However, no anomalies of attenuation have been detected for the mode related to the c 33 elastic modulus, in contradiction to the theoretical predictions based on the previous method, worked out for cubic crystals. In the new method we obtained direct relations between the deformations, related to the crystal moduli, and the local JT modes, calculated the partition functions for each of the three possible JTE problems for systems with an electronic T term, T⊗e, T⊗t 2 and T⊗(e + t 2 ) revealed how these deformations alter the vibronic energy levels responsible for the relaxations in the JT centers. It emerged that in the wurtzite crystal under consideration, only in the T⊗e problem the deformation related to the elastic moduli c 33 displaces all the vibronic energy level uniformly, without relaxation possibilities, thus supporting the new method and explaining the experimental observations. © Published under licence by IOP Publishing Ltd. This research work was carried out within the Russian state assignment No.AAAA-A18-118020190098-5. We acknowledge the support from the HLD at HZDR, member of the European Magnetic Field Laboratory (EMFL), from Russian Foundation for Basic Research (project 18-02-00332 a), and from UrFU Center of Excellence "Radiation and Nuclear Technologies" (Competitiveness Enhancement Program). N.S.A. thanks the Foundation for the Advancement of Theoretical Physics and Mathematics "BASIS" (Russia).

Optical Character Recognition (OCR) is the process of recognizing characters automatically from scanned or image documents. OCR software uses machine learning to recognize characters in the document. Such software needs to pass a training phase to learn how to recognize the letters in the text. In order to implement the training phase the OCR needs to use a standard dataset. The dataset can be used to evaluate the obtained results. In this research, we propose an Arabic printed OCR dataset. To the best of our knowledge, there is no Arabic OCR dataset that is available to be used by the research community with its ground truth with a size that is suitable to build a robust Arabic OCR. The proposed dataset is extracted randomly from Wikipedia to have different topics. It consists of 4,587 Arabic articles with a total of 8,994 images.

Introduction DRC is endowed with enormous mining potential, the exploitation of which promises great hope of economic development. However, the various activities carried out during several years of operation have led to negative environmental and social impacts. The exploitation of mineral deposits has had deleterious effects on the biophysical, socio-economic and health aspects of the surrounding populations. Methods Systematic and impartial assessment of mining activities was conducted in terms of environmental impacts, waste management, implementation of environmental control mitigation measures, and emergency plan, according to national regulations and requirements of The World Bank. Field study required visits to the sites in the Provinces of Katanga, Kasai Oriental and Kasai Occidental, collection of samples of water and soil, collection of health data, and interviews with key personnel including representatives of mining companies, miners and surrounding populations. Result Samples of surface water, groundwater and soil were collected and analysed along with surveys on occupational health and safety issues including noise and observations carried out by the group of experts on the basis of literature in the field. The contaminants that were above the standard included Silver, Arsenic, Copper, Molybdenum, Chromium, Zinc, Manganese, Mercury, Potassium and Sulphur, Nitrites and Nitrates, and suspended organic matter. Resulting population displacements had quite disruptive social effects and serious risks of impoverishment on aboriginal families and residents and damage to the environment. Discussion Issues of waste management, liquid effluents, especially acid mine drainage, and the enormous excavation holes were quite noticeable. The analysis revealed exceedance of the current standards of DRC, WHO or Quebec, concerning the quality of surface and ground water, and soil quality. This mission laid the groundwork for an awareness of the dangers that threaten the environment in general, and especially the populations living in the vicinity of explored mining sites along with mitigation measures.

Speech recognition poses some interesting challenges such as varying acoustic conditions, dialects, and articulation at word's boundaries. Large vocabulary speaker-independent continuous speech recognition systems have recently received significant attention. In this paper, we present a survey of Arabic speech recognition that even has more challenges such as the optional diacritization of the Arabic script. Even though Arabic is a live language that is spreading widely throughout a large area, the research devoted to this technology still in the early stages compared to other languages such as English language. In this study, we highlight the progress made so far in Arabic speech recognition field that include corpora, phonemes, language models, acoustic models, and some promising research directions. This survey reveals that the shortage of freely available continuous speech corpora deserves more research attention in this domain. It also shows a need to compile large corpora or a benchmark, as it will be a key factor to promote the Arabic language research for effective human-computer interaction.

Statistical N-grams language models (LMs) have shown to be very effective in natural language processing (NLP), particularly in automatic speech recognition (ASR) and machine translation. In fact, the successful impact of LMs promote to introduce efficient techniques as well as different types models in various linguistic applications. The LMs mainly include two types that are grammars and statistical language models that is also called N-grams. The main difference between grammars and statistical language models is that the statistical language models are based on the estimation of probabilities for words sequences while the grammars usually do not have probabilities. Despite there are many toolkits that are used to create LMs, however, this work employs two well-known language modeling toolkits with focus on the Arabic text. The implementing toolkits include the Carnegie Mellon University (CMU)-Cambridge Language Modeling Toolkit and the Cambridge University Hidden Markov Model Toolkit (HTK) language modeling toolkits. For clarification, we used a small Arabic text corpus to compute the N-grams for 1-gram, 2-gram, and 3-gram. In addition, this paper demonstrates the intermediate steps that are needed to generate the ARPA-format LMs using both toolkits.

Sentiment analysis of social networks data has recently received a great deal of attention. Social networks are characterized by uncommon language that is different when compared with the standard format of the language. Hence, there is a demand for effective methods to analyze the huge volume of the new word variants that quickly and daily show up in the digital and online world. In text classification, vector space model (VSM) is based on the vocabulary list (i.e. the entire training set words) while ignoring the odd words, which leads to partial loss of textual information. To address this challenge, we propose to use each two-neighboring letters of the word as a basic feature unit instead of using the word itself. That is, instead of using words in VSM, we propose a new method that is based on decomposing each word into a sequence of micro-words, each of which has only two consecutive letters. Two data collections were employed to investigate the performance. The data collections include common (i.e. standard form) and uncommon Arabic text (obtained from Instagram). For the common text, we used a corpus that contains 1,500 documents for training and 500 documents for testing. The proposed method was evaluated using latent semantic indexing (LSI) for textual features and cosine similarity measure for classification. The experimental results show promising results as the proposed method correctly classifies the testing set documents with an accuracy up to 83.6%.