Search

Sentiment analysis in the Arabic language is challenging because of its linguistic complexity. Arabic is complex in words, paragraphs, and sentence structure. Moreover, most Arabic documents contain multiple dialects, writing alphabets, and styles (e.g., Franco-Arab). Nevertheless, fine-tuned bidirectional encoder representations from transformers (BERT) models can provide a reasonable prediction accuracy for Arabic sentiment classification tasks. This paper presents a fine-tuning approach for BERT models for classifying Arabic sentiments. It uses Arabic BERT pre-trained models and tokenizers and includes three stages. The first stage is text preprocessing and data cleaning. The second stage uses transfer-learning of the pre-trained models’ weights and trains all encoder layers. The third stage uses a fully connected layer and a drop-out layer for classification. We tested our fine-tuned models on five different datasets that contain reviews in Arabic with different dialects and compared the results to 11 state-of-the-art models. The experiment results show that our models provide better prediction accuracy than our competitors. We show that the choice of the pre-trained BERT model and the tokenizer type improves the accuracy of Arabic sentiment classification.

Image usage over the internet becomes more and more important each day. Over 3 billion images are shared each day over the internet which raise a concern about how to protect images copyrights? Or how to utilize image sharing experience? This paper proposes a new robust image watermarking algorithm based on compressed sensing (CS) and quantization index modulation (QIM) watermark embedding. The algorithm capitalizes on the CS to compress and encrypt images jointly with Entropy Coding, Arnold Cat Map, Pseudo-random numbers and Advanced Encryption Standard (AES). Our proposed algorithm works under the JPEG standard umbrella. Watermark embedding is done in 3 different locations inside the image using QIM. Those locations differ with each 8-by-8 image block. Choosing which combination of coefficients to be used in QIM watermark embedding depends on selecting a combination from combinations table, which is generated at the same time with projection matrices using a 10-digits Pseudorandom number secret key SK1. After quantization phase, the algorithm shuffles image blocks using Arnold’s Cat Map with a 10-digits Pseudo-random number secret key SK2, followed by a unique method for splitting every 8x8 block into two unequal parts. Part number one will act as the host for two QIM watermarks then goes through encoding phase using Run-Length Encoding (RLE) followed by Huffman Encoding, while part number two goes through sparse watermark embedding followed by a third QIM watermark embedding and compression phase using CS, then Huffman encoder is used to encode this part. The algorithm aims to combine image watermarking, compression and encryption capabilities in one algorithm while balancing how those capabilities works with each other to achieve significant improvement in terms of image watermarking, compression and encryption. 15 different images usually used in image processing benchmarking were used for testing the algorithm capabilities and experiments show that our proposed algorithm achieves robust watermarking jointly with encryption and compression under the JPEG standard framework.

Word Sense Disambiguation (WSD) aims to predict the correct sense of a word given its context. This problem is of extreme importance in Arabic, as written words can be highly ambiguous, 43% of diacritized words have multiple interpretations and the percentage increases to 72% for non-diacritized words. Nevertheless, most Arabic written text does not have diacritical marks. Gloss-based WSD methods measure the semantic similarity or the overlap between the context of a target word that needs to be disambiguated and the dictionary definition of that word (gloss of the word). Arabic gloss WSD suffers from a lack of context-gloss datasets. In this paper, we present an Arabic gloss-based WSD technique. We utilize the celebrated Bidirectional Encoder Representation from Transformers (BERT) to build two models that can efficiently perform Arabic WSD. These models can be trained with few training samples since they utilize BERT models that were pretrained on a large Arabic corpus. Our experimental results show that our models outperform two of the most recent gloss-based WSDs when we test them against the same test data used to evaluate our model. Additionally, our model achieves an F1-score of 89% compared to the best-reported F1-score of 85% for knowledge-based Arabic WSD. Another contribution of this paper is introducing a context-gloss benchmark that may help to overcome the lack of a standardized benchmark for Arabic gloss-based WSD.

With the development of the new era of technology and social networks, forums, blogs, and online sales, combined with the increase of Arabs expressing their opinions, it became an urgent matter to do this research. This paper investigates different deep neural network architectures to perform Arabic sentiment classification combined with word embedding approaches. The models used are Recurrent Neural network (RNN), Bidirectional multi-layer long short-term memory (LSTM), and FastText. Different hyperparameters are used to train each model. In addition, a neural network of Multi-Layer Bidirectional Long Short-Term Memory trained on top of Glove Arabic word embedding with 1.75 billion tokens and 1.5 million words beat the existing methods.

Electrocardiogram (ECG) signal analysis is a critical task in diagnosing the presence of any cardiac disorder. There are limited studies on detecting P-waves in various atrial arrhythmias, such as atrial fibrillation (AFIB), atrial flutter, junctional rhythm, and other arrhythmias due to P-wave variability and absence in various cases. Thus, there is a growing need to develop an efficient automated algorithm that annotates a 2D printed version of P-waves in the well-known ECG signal databases for validation purposes. To our knowledge, no one has annotated P-waves in the MIT-BIH atrial fibrillation database. Therefore, it is a challenge to manually annotate P-waves in the MIT-BIH AF database and to develop an automated algorithm to detect the absence and presence of different shapes of P-waves. In this paper, we present the manual annotation of P-waves in the well-known MIT-BIH AF database with the aid of a cardiologist. In addition, we provide an automatic P-wave segmentation for the same database using a fully convolutional neural network model (U-Net). This algorithm works on 2D imagery of printed ECG signals, as this type of imagery is the most commonly used in developing countries. The proposed automatic P-wave detection method obtained an accuracy and sensitivity of 98.56% and 98.78%, respectively, over the first 5 min of the second lead of the MIT-BIH AF database (a total of 8280 beats). Moreover, the proposed method is validated using the well-known automatically and manually annotated QT database (a total of 11,201 and 3194 automatically and manually annotated beats, respectively). This results in accuracies of 98.98 and 98.9%, and sensitivities of 98.97 and 97.24% for the automatically and manually annotated QT databases, respectively. Thus, these results indicate that the proposed automatic method can be used for analyzing long-printed ECG signals on mobile battery-driven devices using only images of the ECG signals, without the need for a cardiologist.

The goal of this paper is twofold. Firstly, a Siamese-twin random projection neural network (ST-RPNN) is proposed for unsupervised binary hashing of images and compared with state-of-the-art techniques. Secondly, a comparison between Hamming-distance-based retrieval and a proposed bagging trees retrieval (BT-retrieval) algorithm operating directly on the PCA features is made with respect to performance, storage and search time. The ST-RPNN is made of two identical random projection neural networks and is trained to produce similar binary codes for similar input image pairs and different binary codes otherwise. The learning process is divided into two steps: a fast sparse neurons selection algorithm followed by an unsupervised bagging trees algorithm to extract the compact required-length code. Moreover, a BT-retrieval algorithm is proposed in this paper as a fast retrieval tool that ranks the database with respect to a query without distance calculations. Furthermore, (BT-PCA) is a novel extension where the BT-retrieval is applied directly on the PCA features with a significantly lower time search than Hamming-distance-based approach. The proposed technique is compared with ten unsupervised image binary hashing techniques on the COREL1K dataset and the CIFAR10 dataset. The proposed technique obtained better precision–recall results than all compared techniques on the COREL1K dataset, and better than eight of them on the CIFAR10 dataset.

Sparse coding is based on the concept of having a large dictionary of candidate basis vectors. Any given vector is expressed as a sparse linear combination of the dictionary vectors. It has been developed in the signal processing field, and has many applications in data compression and image processing. In this paper we propose applying sparse coding to the time series forecasting field. Specifically, the paper investigates different dictionary based local learning techniques for building predictive models for the time series forecasting problem. The proposed methodology is based on a local learning framework whereby the query point is embedded and coded in terms of a sparse combination of the training dictionary atoms (vectors). Then this embedding is used for estimating the target value of the query point, by applying the same embedding to the target vectors of the dictionary training atoms. We present an experimental study of several sparse coding algorithms. Experiments are performed on the large monthly time series benchmark from the M3 competition, and these experiments showed that the sparse methods Lasso and Elastic-Net presented the best results among the sparse coding algorithms. Moreover, they outperformed the K-nearest neighbor (KNN) regression and most of the compared machine learning and statistical forecasting techniques, especially for higher horizons.

Due to the huge number of online videos uploaded and viewed every day, there is an emerging need nowadays for the action recognition techniques. Applying these techniques in uncontrolled and realistic videos is still a challenging task, considering the large variations in camera motion, viewpoint, cluttered background etc. Moreover, they need to be automated to be able to handle such an amount of different actions. The goal of this study is to introduce a new technique for mining mid-level discriminative patches from videos. These patches are the most representative parts that can describe an action. To achieve this goal, the authors generalise a technique borrowed from 2D images to generate bounding boxes with a high motion and appearance saliencies. Then, a clustering-classification iterative procedure is applied on the generated boxes. Then, they calculate a discriminative score for each box. Finally, they select top ranked boxes to train exemplar-SVM on low-level features which are extracted from the selected boxes. The proposed approach has been evaluated using two challenging datasets YouTube and JHMDB. The experimental results demonstrated the effectiveness of their approach to achieve a better average recognition accuracy than the state-of-the-art techniques.

Fully homomorphic encryption (FHE) technology is a method of encrypting data that allows arbitrary calculations to be computed. Machine learning (ML) and many other applications are relevant to FHE such as Cloud Computing, Secure Multi-Party, and Data Aggregation. Only the authenticated user has the authority to decrypt the ciphertext and understand its meaning, as encrypted data can be computed and processed to produce an encrypted output. Homomorphic encryption uses arithmetic circuits that focus on addition and multiplication, allowing the user to add and multiply integers while encrypted. This paper discusses the performance of the Brakerski-Fan-Vercauteren scheme (BFV) and Cheon, Kim, Kim, and Song (CKKS) scheme using one of the most important libraries of FHE “Microsoft SEAL”, by applying certain arithmetic operations and observing the time consumed for every function applied in each scheme and the noise budget after every operation. The results obtained show the difference between the two schemes when applying the same operation and the number of sequential operations each can handle.

A novel privacy preserving classification approach is proposed in this paper which employs multiple encrypted random forests with two compressed sensing encryption levels. At the first level, each feature vector is CS-encrypted using a different random matrix for each forest. At query time, the user selects one matrix randomly from a set of R different matrices, and gets R encrypted results of which only one will be used. At the second level, the class-label information at each tree leaf is encrypted using a different CS matrix for each tree. During recognition, the cloud adds all the encrypted leaf CS vectors for each of the R forests and sends to the user, where only one of them goes through sparse recovery to find the class label. Experiments on COREL1K and CIFAR10 image classification tasks show that the proposed approach achieves classification accuracy similar or better than nearest neighbour classifier with plaintext features. Also, correlation-based results show that the multiple-forest approach offers good level of security between the feature vectors while the class-label CS encryption approach achieves privacy to prevent the cloud from knowing the classification outcome.

Embryonic stem cells (ESC) are pluripotent cells, had been obtained from an embryo in a few days old in which all 200+ kinds of tissue in the human body originate a fact which makes them potentially important in medicine. DNA methylation on the other hand plays an important role in gene modification leading to change in biological processes as pluripotent and differentiated cells or reasoned for many diseases. Cytosines in CpG dinucleotides can be methylated leading to concentrate to predict methylation in CpG loci inside DNA by helping of results coming from whole genome bisulfite sequencing (WGBS) experiment but very expensive and Illumina 450K array experiment that covers less than 2% of CpG locus. A machine learning model has been implemented using support vector machine classifier (SVM) that is trained on chromosome 18 of the human H1 ESC with both 450K array data and WGBS. The SVM gives an accuracy of 99% when testing on chromosome 18 of the human H9 ESC to predict methyl or un-methyl for each CpG loci in the chromosome separately.

In this paper, we propose a joint image compression and encryption technique based on Compressed Sensing (CS). Our work is based on JPEG standard. After the quantization stage, blocks in the DCT domain become sparse, however, if we use CS on the whole block the Entropy coding compression is severely affected. As a result, we split the DCT coefficients between Entropy coding path and CS path followed by its Entropy coder. As an added value for CS, partial encryption is achieved. To split the DCT coefficients to Entropy and CS paths, we developed five different algorithms; each algorithm uses a unique method to split every 8×8 block in the image into two parts. First part is encoded using Huffman coding, while the second part goes through compression stage using CS, then encoded using Huffman coding. At the end, we combine encoded parts to get the compressed and encrypted image. To strengthen the image encryption, we use the CS secret key to shuffle image blocks using Arnold Cat Map. The five algorithms are tested on 15 popular images and the results show that 2 techniques can achieve compression gain over JPEG as well as the partial encryption, all at a PSNR very close to JPEG.

The aim of this paper is to present and discuss issues related to the telematics business and management in Bahrain. Telematics is a part of Information and Communication Technologies (ICT), which plays a major role in the economic development of countries. Telematics applications are becoming increasingly important in modern transportation to increase reliability, safety and security of both vehicles and drivers. There are many applications for telematics; some of these applications are telematics for educations, telematics for health, and vehicle telematics. One of the most important applications of telematics is vehicle telematics, which includes vehicle tracking, fleet management, container tracking, trail tracking. In this paper we are focusing on vehicle telematics business and management in Bahrain market.

In this paper a watermark embedding and recovery technique is proposed based on the compressed sensing framework. Both the watermark and the host signal are sparse, each in its own domain. In recovery, the L1-minimization is used to recover the watermark and the host signal almost perfectly in clean conditions. The proposed technique is tested on MP3 audio compression-decompression attack and additive noise attack. Bit error rates are compared with standard spread spectrum embedding. The proposed technique is implemented for both time domain and frequency domain embedding with a unified approach. The WalshHadamard transform (WHT), the discrete cosine transform (DCT) and the Karhunen-Loeve transform (KLT) are compared in the host signal sparsifying process. Significant performance improvements in all tested conditions are achieved against the spread spectrum embedding. A payload as high as 172bps in additive noise attacks, 86bps in 128kbps MP3 attacks and 11bps in 64kbps MP3 attacks are achieved at small bit error rates and acceptable MP3 audio signal quality.

Due to the huge amount of online videos uploaded and viewed every day, there is an emerging need nowadays for action recognition techniques. These techniques need to consider the large variations in camera motion, viewpoint, cluttered background, etc. Moreover, they need to be unsupervised or weakly supervised to can absorb such amount of different actions. The goal of this paper is to introduce a new unsupervised technique for mining mid-level discriminative patches from videos. These patches are the most representative parts that can describe an action. To achieve this goal, we generalize a technique borrowed from 2D images to generate bounding boxes with high motion and appearance saliencies then apply iterative clustering/classification procedure on generated boxes. Then, we calculate discriminative score for each box and finally select top ranked boxes to train Exemplar-SVM on low-level features extracted inside selected boxes. We evaluate our approach on a challenging dataset namely YouTube and the experimental results demonstrate the effectiveness of our approach to achieve better average recognition accuracy than the state-of-the-art techniques.

The text alignment process is an important activity in the statistical machine translation (SMT) upon the training phase of SMT systems on huge amounts of parallel corpora. We are evaluating five of the popular text aligners and analyzing the methodologies used in each of them, then we are introducing new open source automated test bed for testing those text alig ners and generating reports called pyTester, and finally discuss our observations about the test results generated from pyTester .

The aim of this paper is to present and demonstrate the main challenges and obstacles facing Telematics companies entering the market in Bahrain. Telematics is part of Information and communication technologies (ICT) which plays a major role in the economic development of countries. However, there are many obstacles and challenges facing new businesses that venture into the small market of Bahrain; e.g. the small market size, competition and lack of financing channels (e.g. venture capital).

This paper proposes a dictionary-based L1-norm sparse coding for time series prediction which requires no training phase, and minimal parameter tuning, making it suitable for nonstationary and online prediction applications. The prediction process is formulated as a basis pursuit L1-norm problem, where a sparse set of weights is estimated for each test vector. Constrained sparse coding formulations are compared including sparse local linear embedding and sparse nearest neighbor embedding. 16 time series datasets are used to test the approach for offline time series prediction where the training data is fixed. The proposed approach is also compared to Bagging trees (BT), least-squares support vector regression (LSSVM) and regularized Autoregressive model. The proposed sparse coding prediction shows better performance than the LSSVM that uses 10-fold cross validation and significantly better performance than regularized AR and Bagging trees. In average, a few thousand sparse coding predictions can be done while the LSSVM is training making the proposed technique suitable for online prediction and highly nonstationary data.

In this paper, a novel Commutative Watermarking and Partial Encryption (CWPE) algorithm based on Discrete Wavelet Transform and Discrete Cosine Transform (DWT-DCT) for watermarking and Multi-Map Wavelet Chaotic Encryption (MMW-CE) is proposed. The original host image is first decomposed into four sub-bands using (DWT), each sub-band coefficients are relocated using Arnold transform to create a noiselike version, then apply partial encryption scheme using chaotic scrambled random number pattern bitwise XOR with the scrambled horizontal coefficients only and the shuffled approximation coefficients are divided into non-overlapping and equal sized blocks. Watermark embedding process is based on extracting the (DCT) middle frequencies of the encrypted approximation coefficients blocks. Comparison based threshold of the extracted DCT mid-band coefficients, watermark bits are embedded in the coefficients of the corresponding DCT middle frequencies. The experimental results show that the proposed algorithm is robust against common signal processing attacks. The proposed algorithm is able to reduce encryption to one quarter of the image information. Statistical and differential analyses are performed to estimate the security strength of the proposed algorithm. The results of the security analysis show that the proposed algorithm provides a high security level for real time application.

Online nonstationary time series prediction requires continuous model parameter update to be able to track the changing characteristics of the data. Recent approaches such as sequential parameter adaptation and varying coefficient models require re-estimation of the prediction model parameter with every new available data. An online dictionary-based prediction approach is proposed in this paper where new data is added to the dictionary while maintaining the dictionary size by using 3 different methods. The dictionary based prediction employs a sparse coding model where an L1-norm convex optimization problem is solved for each new time series vector, while no training phase is required. The validity of the proposed approach is tested on 4 nonstationary time series data sets reported in the literature. Results show that the proposed online prediction approach reaches comparable results with more complex techniques which require more computationally demanding parameter adaptation.

Extreme Learning Machines (ELM) is a class of supervised learning models that have three basic steps: A random projection of the input space followed by some nonlinear operation and finally a linear output layer of weights. The basic ELM uses pseudo matrix inverse to estimate the output layer weights which usually leads to over fitting. Recent research suggested the use of L2-norm regularization to enhance the sparsity of the output layer. This paper proposes the use of the L1-norm LASSO formulation, since the L1-norm promotes sparsity in the solution of the output layer weights and it has been shown to produce the sparsest solutions in many applications. Extensive comparison between the basic ELM, the L1-norm and the L2-norm is conducted over a number of classification tasks, with significant improvement in sparseness using the proposed approach with better performance than that reported in the literature.

In this research different clustering techniques are applied for grouping transcribed textual documents obtained out of audio streams. Since audio transcripts are normally highly erroneous, it is essential to reduce the negative impact of errors gained at the speech recognition stage. In attempt to overcome some of these errors, different stemming techniques are applied on the transcribed text. The goal of this research is to achieve automatic topic clustering of transcribed speech documents, and investigate the impact of applying stemming techniques in combination with a Chi-square similarity measure on the accuracy of the selected clustering algorithms. The evaluation—using F-Measure—showed that using root-based stemming in combination of spectral clustering technique achieved the highest accuracy.

Automated document clustering is an important text mining task especially with the rapid growth of the number of online documents present in Arabic language. Text clustering aims to automatically assign the text to a predefined cluster based on linguistic features. This research proposes an enhanced grid based clustering algorithm. The main purpose of this algorithm is to divide the data space into clusters with arbitrary shape. These clusters are considered as dense regions of points in the data space that are separated by regions of low density representing noise. Also it deals with making clustering the data set with multi-densities and assigning noise and outliers to the closest category. This will reduce the time complexity. Unclassified documents are preprocessed by removing stops words and extracting word root used to reduce the dimensionality of feature vectors of documents. Each document is then represented as a vector of words and their frequencies. The accuracy is presented according to time consumption and the percentage of successfully clustered instances. The results of the experiments that were carried out on an in-house collected Arabic text have proven its effectiveness of the enhanced clustering algorithm with average accuracy 89 %.

the aim of this paper is to introduce a new technique that enhances online translation from English to Arabic for a specific domain. This enhancement is achieved by training a new "Arabic online engine translation" to "Arabic manual translation" model that corrects common errors in the online translation. This paper focuses on two popular online translation engines which are Google and Bing. The translation enhancement is done in two steps. First step is training the translation of each online engine on Moses toolkit separately to generate translation model that corrects the output of online translation engine, the result of this step is corrected Google translation and corrected Bing translation which is better than the original translation. Second step is choosing one sentence from the two corrected sentence using a language model trained on the manual Arabic translation which gives more enhancement than step 1.Two metrics are used to evaluate proposed model enhancement, BLEU Score (Bilingual Evaluation Understudy) and NIST Score (National Institute of Standards and Technology). The proposed model has obtained 32.44% BLEU Score over Google and 40.91% BLEU Score over Bing for LDC2004T14 dataset, 18.08% BLEU Score over Google and 26.55% BLEU Score over Bing for Distinct LDC2004T14 dataset, and 10.69% BLEU Score over Google and 9.94% BLEU Score over Bing for MEDAR dataset.

Model order reduction (MOR) is an approximation approach where a high order complex system is modeled by a low order parametric system. Power system transients, RLC interconnects in deep submicron technology, and large scale dynamical systems are just a few examples of MOR practical applications. In this paper, the focus is on approximating the impulse response of high order RLC-like systems. In the proposed approach an overcomplete dictionary of damped sinusoids is constructed and the sparse coding paradigm is used to find the sparsest solution to the given impulse response. Two modifications of the basic dictionary with added random atoms are used. A particle swarm optimization with constraints algorithm is employed to refine the sparse coding model, and finally, the best model is chosen adaptively. A system of order 24 is simulated and reduced models of orders 4, 6 and 8 are tested for different dictionary sizes.

In this paper, a novel Commutative Watermarking and Partial Encryption (CWPE) algorithm using single level 2-Dimension Discrete Wavelet Transform (CWPE 2D DWT) and Multi-Map Orbit Hopping Chaotic System (MMOH-CS) is proposed. The experimental results show that the proposed algorithm is robust against common signal processing attacks such as Salt & Pepper noise, Gaussian noise, Wiener filter, JPEG compression, Resizing and Cropping. The proposed scheme is able to reduce encryption to one quarter of the image information. Statistical analysis, differential analysis and key sensitivity test are performed to estimate the security strength of the proposed algorithm. The results of the security analysis show that the proposed algorithm provides a high security level for real time application.