Ahmad Al-Rubaie

Semi-Supervised Latent Dirichlet Allocation and Its Application for Document Classification

Latent Dirichlet Allocation (LDA) is an unsupervised topic modeling method widely applied in natural language processing. However, standard LDA does not permit the use of supervised labels to incorporate expert knowledge into the learning procedure. This paper describes a semi-supervised LDA (ssLDA) method that supports multiple-topic labels per document, to incorporate available expert knowledge during the model construction. This improvement enables the alignment of resulting model with human expectations for topic modeling and extraction. We apply ssLDA to document classification problem on benchmark datasets. We investigate and compare how the size of training set and proportion of supervised data affect the final model structure and improve the prediction accuracy.

Real time road traffic monitoring alert based on incremental learning from tweets

Social media has become an important source of near-instantaneous information about events and is increasingly also being analysed to provide predictive models, sentiment analysis and so on. One domain where social media data has value is transport and this paper looks at the exploitation of Twitter data in traffic management. A key issue is the identification and analysis of traffic-relevant content. A smart system is needed to identify traffic related tweets for traffic incident alerting. This paper proposes an instant traffic alert and warning system based on a novel LDA-based approach (“tweet-LDA”) for classification of traffic-related tweets. The system is evaluated and shown to perform better than related approaches.

Incremental learning with partial-supervision based on hierarchical Dirichlet process and the application for document classification

Partial-supervision: use available knowledge to guide model learning process for better accuracy.Incremental learning: adjust parameters and model structure to the latest information.Introduce granular computing idea to achieve better accuracy and detect new emergent categories. Hierarchical Dirichlet process (HDP) is an unsupervised method which has been widely used for topic extraction and document clustering problems. One advantage of HDP is that it has an inherent mechanism to determine the total number of clusters/topics. However, HDP has three weaknesses: (1) there is no mechanism to use known labels or incorporate expert knowledge into the learning procedure, thus precluding users from directing the learning and making the final results incomprehensible; (2) it cannot detect the categories expected by applications without expert guidance; (3) it does not automatically adjust the model parameters and structure in a changing environment. To address these weaknesses, this paper proposes an incremental learning method, with partial supervision for HDP, which enables the topic model (initially guided by partial knowledge) to incrementally adapt to the latest available information. An important contribution of this work is the application of granular computing to HDP for partial-supervision and incremental learning which results in a more controllable and interpretable model structure. These enhancements provide a more flexible approach with expert guidance for the model learning and hence results in better prediction accuracy and interpretability.

Real-Time Traffic Event Detection From Social Media

Smart communities are composed of groups, organizations, and individuals who share information and make use of that shared information for better decision making. Shared information can come from many sources, particularly, but not exclusively, from sensors and social media. Social media has become an important source of near-instantaneous user-generated information that can be shared and analyzed to support better decision making. One domain where social media data can add value is transportation and traffic management. This article looks at the exploitation of Twitter data in the traffic reporting domain. A key challenge is how to identify relevant information from a huge amount of user-generated data and then analyze the relevant data for automatic geocoded incident detection. The article proposes an instant traffic alert and warning system based on a novel latent Dirichlet allocation (LDA) approach (“tweet-LDA”). The system is evaluated and shown to perform better than related approaches.

Enabling the Next Generation Learning Environment

As more and more services that universities traditionally offer are provided through or underpinned by systems made of computers and networks, the role of these systems become more important. Previously independent functions can now be combined, or easily exchange data and information, hence offering new opportunities. In order to leverage and exploit these new possibilities, these services must be built to access each others data and functionalities in an open, secure and simple manner. This paper describes a smart learning platform that provides the backbone to the University of the Future for the next generation campus environment.

Automatic Arabic Summarization: A survey of methodologies and systems

Abstract Text summarization has been a field of intensive research over the last 50 years, especially for commonly-used and relatively simple-grammar languages such as English. Moreover, the unprecedented growth in the amount of online information available in many languages to users and businesses, including news articles and social media, has made it difficult and time consuming for users to identify and consume sought after content. Hence, an automatic text summarization for various languages to generate accurate and relevant summaries from the huge amount of information available is essential nowadays. Techniques and methodologies for Arabic text summarization are still immature due to the inherent complexity of the Arabic language in terms of both structure and morphology. This paper describes the main challenges for Arabic text summarization and surveys the various methodologies and systems in the literature. This survey would be a good basis for the design of an Arabic automatic text summarization that combines the various “good” features of the existing systems and dismiss the “not-so-good” features.

Smart text-classification of user-generated data in educational social networks

Nowadays, face-to-face interpersonal communication has been gradually replaced by communications via virtual social network platforms, which applies to the new generation of education. The amount of user-generated data in social networking sites is increasing day by day. Understanding and consuming this great amount of data has become a harder task. Classifying the user-generated data (mainly text) can help simplify the user experience by providing them dynamic personalized recommender. Filtering the data, and providing users with what is relevant to them, will help them utilize this data more effectively. In education, recommending relevant learning content to learners in educational social networking sites saves them the arduous task of sifting through a huge amount of information. This paper introduces the partial-supervised learning for Hierarchical Dirichlet Process (HDP) for text classification with inherent hierarchical structure in education. This enables the use of partial known model structure and labels as expert knowledge to guide the model learning procedure from the text without labels. Compared with the existing partial/semi-supervised HDP, the proposed method is able to make use of the known labels for not only structure construction, but also parameter learning. This enhancement provides a more flexible way and better guide for the model learning from the unlabelled documents. We experimentally investigate the contribution of partial knowledge to guide the model learning process. The proposed partial-supervision for HDP is applied to student micro-blog automatic classification and adds intelligence to our student social media platform (ELSE).