Hazem M. Raafat

Semantic Query Expansion for Arabic Information Retrieval

Traditional keyword based search is found to have some limitations. Such as word sense ambiguity, and the query intent ambiguity which can hurt the precision. Semantic search uses the contextual meaning of terms in addition to the semantic matching techniques in order to overcome these limitations. This paper introduces a query expansion approach using an ontology built from Wikipedia pages in addition to other thesaurus to improve search accuracy for Arabic language. Our approach outperformed the traditional keyword based approach in terms of both F-score and NDCG measures.

Word Representations in Vector Space and their Applications for Arabic

A lot of work has been done to give the individual words of a certain language adequate representations in vector space so that these representations capture semantic and syntactic properties of the language. In this paper, we compare different techniques to build vectorized space representations for Arabic, and test these models via intrinsic and extrinsic evaluations. Intrinsic evaluation assesses the quality of models using benchmark semantic and syntactic dataset, while extrinsic evaluation assesses the quality of models by their impact on two Natural Language Processing applications: Information retrieval and Short Answer Grading. Finally, we map the Arabic vector space to the English counterpart using Cosine error regression neural network and show that it outperforms standard mean square error regression neural networks in this task.

Deep learning framework with confused sub-set resolution architecture for automatic Arabic diacritization

The Arabic language belongs to a group of languages that require diacritization over their characters. Modern Standard Arabic (MSA) transcripts omit the diacritics, which are essential for many machine learning tasks like Text-To-Speech (TTS) systems. In this work Arabic diacritics restoration is tackled under a deep learning framework that includes the Confused Sub-set Resolution (CSR) method to improve the classification accuracy, in addition to an Arabic Part-of-Speech (PoS) tagging framework using deep neural nets. Special focus is given to syntactic diacritization, which still suffers low accuracy as indicated in prior works. Evaluation is done versus state-of-the-art systems reported in literature, with quite challenging datasets collected from different domains. Standard datasets like the LDC Arabic Tree Bank are used in addition to custom ones we have made available online to allow other researchers to replicate these results. Results show significant improvement of the proposed techniques over other approaches, reducing the syntactic classification error to 9.9% and morphological classification error to 3% compared to 12.7% and 3.8% of the best reported results in literature, improving the error by 22% over the best reported systems.

A tree structured neural network

A tree structured system for pattern classification is proposed. It uses the feedforward neural network with back-propagation (FN) as a building block. A single FN is used to classify all of the given patterns, then a confusion matrix is carefully studied and used to divide the patterns into groups. This process is repeated by training new FNs with these groups then dividing them into subgroups and so on, until no more grouping could be obtained. It is shown that by this approach, the available feature set can be used more effectively. The testing environment of this work is the isolated handwritten Arabic character set, which is a problem of reasonable complexity. However, the suggested method can be applied to other pattern classification problems. Dividing a large problem into smaller and easier ones is the target that is successful reached.<<ETX>>

Automatic Arabic diacritics restoration based on deep nets

In this paper, Arabic diacritics restoration problem is tackled under the deep learning framework presenting Confused Subset Resolution (CSR) method to improve the classification accuracy, in addition to Arabic Part-of-Speech (PoS) tagging framework using deep neural nets. Special focus is given to syntactic diacritization, which still suffer low accuracy as indicated by related works. Evaluation is done versus state-of-the-art systems reported in literature, with quite challenging datasets, collected from different domains. Standard datasets like LDC Arabic Tree Bank is used in addition to custom ones available online for results replication. Results show significant improvement of the proposed techniques over other approaches, reducing the syntactic classification error to 9.9% and morphological classification error to 3% compared to 12.7% and 3.8% of the best reported results in literature, improving the error by 22% over the best reported systems

Fog Intelligence for Real-Time IoT Sensor Data Analytics

The evolution of the Internet of things and the continuing increase in the number of sensors connected to the Internet impose big challenges regarding the management of the resulting deluge of data and network latency. Uploading sensor data over the web does not add value. Therefore, an efficient knowledge extraction technique is badly needed to reduce the amount of data transfer and to help simplify the process of knowledge management. Homoscedasticity and statistical features extraction are introduced in this paper as novelty detection enabling techniques, which help extract the important events in sensor data in real time when used with neural classifiers. Experiments have been conducted on a fog computing platform. System performance has been also evaluated on an occupancy data set and showed promising results.

A novel training weighted ensemble (TWE) with application to face recognition

Individual classifiers that are fully trained are unstable especially when the database conditions are changed. Moreover, designing a unique classifier with the suitable parameters to achieve acceptable performance is a non-trivial task. Combined classifiers, which consist of a set of individually trained classifiers, are introduced to avoid the previous problems. There are two key issues in the combination of classifiers. The first issue is how to obtain the set of base classifiers to combine. The second issue is how to fuse the decisions of those classifiers. In this paper, weak Learning Vector Quantization (LVQ) neural networks have been used as base classifiers. Also, a new combination technique which is based on training-weighted voting is introduced. Other factors that greatly affect the performance of a combined classifier are related to the type of the individual classifiers, the training parameters, database size and nature, etc. These factors have been considered in the design of the proposed combined classifier. TWE has been experimentally tested on five standard face databases: Yale, ORL, Grimace, Faces94 and Faces95 and has demonstrated excellent performance. Analysis of the ensemble stability has shown promising results.

Lattice-based ranking for service trust behaviors

In service-based communities, services strive to collaborate to successfully perform their tasks. In this context, trust is becoming a paramount factor in deciding whether or not to interact with a service. In this paper, we deal with trust from the perspective of behavior analysis and propose a novel feature-based approach for the ranking of trust behaviors. We represent the trust behavior of a service by a sequence of trust observations collected over a certain time frame, called the trust sequence. Our approach spans over a lattice-based trust ranking algorithm for trust sequences, called LattRank. LattRank is inspired by the top-down breadth-first traversal of an attribute-value lattice. It assigns to each trust sequence a ranking score based on the trust order constraints. The ranking score indicates to which extent a service is stable in the trustworthy behavior. We prove formally the correctness of LattRank and evaluate it using ranking evaluation metrics through a comprehensive experimental study.

Automated visual inspection of flat surface products using feature fusion

Defect detection on industrial flat surface products like textiles, steel slabs, metal plates, plastic films, painted car body, parquet slabs and paper is a necessary requirement for quality control and satisfaction of consumers. This paper presents a system for feature extraction and fusion in order to enhance the performance of the defect detection process. A multi-feature fusion technique based on PCA is presented. Features based on Co-occurrence matrix, Laws filters, moment invariants, moment of inertia and standard deviation of gray levels are integrated into a one dimensional feature vector which uniquely differentiates the normal from abnormal textures of a flat surface product. PCA has been used to reduce the feature set into eight significant features. A learning vector quantization neural network is used for classification of product surface image blocks as normal or abnormal. Detection accuracies using the individual feature sets and the fused features are compared. The results obtained from multi-feature fusion outperformed those obtained from the individual feature sets and indicate that the multi-feature fusion improves the accuracy of detection and speeds up the process. Empirical results show the high accuracy of the presented approach (97.96%).

Committee machines for facial-gender recognition

This paper presents a new approach for building a committee machine (LVQCM) that is based on learning vector quantization (LVQ) neural networks. The proposed committee machine was then applied to solve the problem of facial gender recognition. Design of individual classifiers is time consuming and results in inaccurate and unstable classifiers. Settling on the right design parameters of a classifier is a non-trivial task. To avoid the abovementioned problems, a committee machine is implemented. Experimental results based on Kuwait University and Stanford University face databases indicate that the performance of the proposed committee machine (99.02%) outperforms that of the best individual classifier used in that combination (93%). Majority voting is used for combining the individual decisions of a group of LVQ weak classifiers generated and trained under different conditions. The experimental results also show that LVQCM outperforms other recently published methods such as: the K-Means, 2