Search

The Web of Linked Open Data (LOD) is a decentralized effort in publishing datasets using a set of conventions to make them accesssible, notably thought RDF and SPARQL. Links across nodes in published datasets are thus critical in getting value for the LOD cloud as a collective effort. Connectivity among the datasets can occur through these links. Equivalence relationship is one of the fundamental links that connects different schemas or datasets, and is used to assert either class or instance equivalence. In this article, we report an empirical study on the equivalences found in over 59 million triples from datasets accessible via SPARQL endpoints in open source data portals. Metrics from graph analysis have been used to examine the relationships between repositories and determine their relative importance as well as their ability to facilitate knowledge discovery.

This paper contributes to compiling and publishing a structured dataset from the scientific literature of the Linked Data on the Web (LDOW) workshop series. This workshop was the primary venue for publishing the frontier topics related to Semantic Web, Linked Open Data, and Web of Data for around ten years. Further, we analyze this dataset using a scientometric approach and n-gram analysis. We apply an n-gram analysis of all the presented titles and abstracts of the dataset to learn the evolving topics discourse. Our analysis reveals interesting patterns over the evolution of the concerning topics in the community of Linked Open Data. We publish this dataset online 3 under GNU General Public License v3.0.

Within the Open Discovery Space our study (T.8.4) focused on ”Enhanced Social Data Visualization & Navigation Services. This deliverable provides the prototype report regarding the deployment of adapted visualization and navigation services to be integrated in the ODS Social Data Management Layer.

With the proliferation of educational data on the Web, publishing and interlinking eLearning resources have become an important issue nowadays. Educational resources are exposed under heterogeneous Intellectual Property Rights (IPRs) in different times and formats. Some resources are implicitly related to each other or to the interest, cultural and technical environment of learners. Linking educational resources to useful knowledge on the Web improves resource seeking. This becomes crucial for moving from current isolated eLearning repositories towards an open discovery space, including distributed resources irrespective of their geographic and system boundaries. Linking resources is also useful for enriching educational content, as it provides a richer context and other related information to both educators and learners. On the other hand, the emergence of the so-called "Linked Data" brings new opportunities for interconnecting different kinds of resources on the Web of Data. Using the Linked Data approach, data providers can publish structured data and establish typed links between them from various sources. To this aim, many tools, approaches and frameworks have been built to first expose the data as Linked Data formats and to second discover the similarities between entities in the datasets. The research carried out for this PhD thesis assesses the possibilities of applying the Linked Open Data paradigm to the enrichment of educational resources. Generally speaking, we discuss the interlinking educational objects and eLearning resources on the Web of Data focusing on existing schemas and tools. The main goals of this thesis are thus to cover the following aspects: -- Exposing the educational (meta)data schemas and particularly IEEE LOM as Linked Data -- Evaluating currently available interlinking tools in the Linked Data context -- Analyzing datasets in the Linked Open Data cloud, to discover appropriate datasets for interlinking -- Discussing the benefits of interlinking educational (meta)data in practice.

A Thesis Submitted to the Faculty of Graduate Studies and Research In Partial Fulfillment of the Requirements for the Degree of Master of Science in Computer Science, University of Regina. xviii, 109 p. The advancement in Electronic Health Records (EHR) and clinical prediction has led to research on effectively predicting patients’ health conditions and events to assist medical professionals. A Delay of appropriate medical attention would result in the loss of life. Prediction of early mortality, length of stay, and acute respiratory failure based on vital measurements such as temperatures, blood pressure, heart rate, sugar level, etc., would assist in saving life and cost. EHR records are sparse and noisy; hence patients have varied vital measurements and length of stay. Due to the lack of a unified digital health record-keeping system, high sparsity, noise, and irregularity exist in the Medical Information Mart for Intensive Care - III (MIMIC III) data. One of the most challenging tasks in clinical prediction is data missingness and data variance. In addition, the clinical data has properties such as homeostasis and diminishing effects of vital measurements. In such a case, the traditional data imputation techniques are inappropriate. We introduced a novel deep learning model that consists of a decay mechanism, modular recurrent networks, and a custom loss function that learns minor classes i.e., positive classes. The decay mechanism enables learning from patterns in sparse data. The modular network allows multiple recurrent networks to pick only relevant input based on the attention score at a given time-stamp. The custom class balance loss function is responsible for learning minor classes based on samples provided in a training data set. In our experiment, we have used identical data sets for all modi els. Our proposed model outperformed the state-of-the-art models such as Gated Recurrent Unit (GRU), Long-Short Term Memory (LSTM), Gated Recurrent Unit Decay (GRUD), and Recurrent Independent Mechanism (RIM) on F1-Score, Area under Receiver Operating Curve (AUROC), and Precision Recall Area under Curve (PRAUC).In this project, we presented a novel method for clinical prediction called Recurrent Independent Mechanism Decay (RIM-Decay). We took vital clinical measurements at different time intervals as input and custom loss functions to handle highly imbalanced data for mortality prediction and length of stay. Student yes