Michalis Mountantonakis

Quantifying the Connectivity of a Semantic Warehouse and Understanding its Evolution over Time

In many applications one has to fetch and assemble pieces of information coming from more than one source for building a semantic warehouse offering more advanced query capabilities. In this paper the authors describe the corresponding requirements and challenges, and they focus on the aspects of quality and value of the warehouse. For this reason they introduce various metrics (or measures) for quantifying its connectivity, and consequently its ability to answer complex queries. The authors demonstrate the behaviour of these metrics in the context of a real and operational semantic warehouse, as well as on synthetically produced warehouses. The proposed metrics allow someone to get an overview of the contribution (to the warehouse) of each source and to quantify the value of the entire warehouse. Consequently, these metrics can be used for advancing data/endpoint profiling and for this reason the authors use an extension of VoID (for making them publishable). Such descriptions can be exploited for dataset/endpoint selection in the context of federated search. In addition, the authors show how the metrics can be used for monitoring a semantic warehouse after each reconstruction reducing thereby the cost of quality checking, as well as for understanding its evolution over time.

On measuring the lattice of commonalities among several linked datasets

A big number of datasets has been published according to the principles of Linked Data and this number keeps increasing. Although the ultimate objective is linking and integration, it is not currently evident how connected the current LOD cloud is. Measurements (and indexes) that involve more than two datasets are not available although they are important: (a) for obtaining complete information about one particular URI (or set of URIs) with provenance (b) for aiding dataset discovery and selection, (c) for assessing the connectivity between any set of datasets for quality checking and for monitoring their evolution over time, (d) for constructing visualizations that provide more informative overviews. Since it would be prohibitively expensive to perform all these measurements in a naive way, in this paper we introduce indexes (and their construction algorithms) that can speedup such tasks. In brief, we introduce (i) a namespace-based prefix index, (ii) a sameAs catalog for computing the symmetric and transitive closure of the owl:sameAs relationships encountered in the datasets, (iii) a semantics-aware element index (that exploits the aforementioned indexes), and finally (iv) two lattice-based incremental algorithms for speeding up the computation of the intersection of URIs of any set of datasets. We discuss the speedup obtained by the introduced indexes and algorithms through comparative results and finally we report measurements about connectivity of the LOD cloud that have never been carried out so far.

MatWare: Constructing and Exploiting Domain Specific Warehouses by Aggregating Semantic Data

In many applications one has to fetch and assemble pieces of information coming from more than one web sources such as SPARQL endpoints. In this paper we describe the corresponding requirements and challenges, based on our experience, and then we present a process and a tool that we have developed, called MatWare , for constructing such semantic warehouses. We focus on domain-specific warehouses, where the focus is given on the aspects of scope control, connectivity assessment, provenance, and freshness. MatWare (Materialized Warehouse) is a tool that automates the construction (and reconstruction) of such warehouses, and offers methods for tackling the aforementioned requirements. Finally we report our experiences from using it for building, maintaining and evolving an operational semantic warehouse for the marine domain, that is currently in use by several applications ranging from e-infrastructure services to smart phone applications.

Scalable Methods for Measuring the Connectivity and Quality of Large Numbers of Linked Datasets

Although the ultimate objective of Linked Data is linking and integration, it is not currently evident how connected the current Linked Open Data (LOD) cloud is. In this article, we focus on methods, supported by special indexes and algorithms, for performing measurements related to the connectivity of more than two datasets that are useful in various tasks including (a) Dataset Discovery and Selection; (b) Object Coreference, i.e., for obtaining complete information about a set of entities, including provenance information; (c) Data Quality Assessment and Improvement, i.e., for assessing the connectivity between any set of datasets and monitoring their evolution over time, as well as for estimating data veracity; (d) Dataset Visualizations; and various other tasks. Since it would be prohibitively expensive to perform all these measurements in a naïve way, in this article, we introduce indexes (and their construction algorithms) that can speed up such tasks. In brief, we introduce (i) a namespace-based prefix index, (ii) a sameAs catalog for computing the symmetric and transitive closure of the owl:sameAs relationships encountered in the datasets, (iii) a semantics-aware element index (that exploits the aforementioned indexes), and, finally, (iv) two lattice-based incremental algorithms for speeding up the computation of the intersection of URIs of any set of datasets. For enhancing scalability, we propose parallel index construction algorithms and parallel lattice-based incremental algorithms, we evaluate the achieved speedup using either a single machine or a cluster of machines, and we provide insights regarding the factors that affect efficiency. Finally, we report measurements about the connectivity of the (billion triples-sized) LOD cloud that have never been carried out so far.

How Linked Data can Aid Machine Learning-Based Tasks

The discovery of useful data for a given problem is of primary importance since data scientists usually spend a lot of time for discovering, collecting and preparing data before using them for various reasons, e.g., for applying or testing machine learning algorithms. In this paper we propose a general method for discovering, creating and selecting, in an easy way, valuable features describing a set of entities for leveraging them in a machine learning context. We demonstrate the feasibility of this approach by introducing a tool (research prototype), called \(\mathtt{LODsyndesis}_\mathcal{ML}\), which is based on Linked Data technologies, that (a) discovers automatically datasets where the entities of interest occur, (b) shows to the user a big number of useful features for these entities, and (c) creates automatically the selected features by sending SPARQL queries. We evaluate this approach by exploiting data from several sources, including British National Library, for creating datasets in order to predict whether a book or a movie is popular or non-popular. Our evaluation contains a 5-fold cross validation and we introduce comparative results for a number of different features and models. The evaluation showed that the additional features did improve the accuracy of prediction.

High Performance Methods for Linked Open Data Connectivity Analytics

The main objective of Linked Data is linking and integration, and a major step for evaluating whether this target has been reached, is to find all the connections among the Linked Open Data (LOD) Cloud datasets. Connectivity among two or more datasets can be achieved through common Entities, Triples, Literals, and Schema Elements, while more connections can occur due to equivalence relationships between URIs, such as owl:sameAs, owl:equivalentProperty and owl:equivalentClass, since many publishers use such equivalence relationships, for declaring that their URIs are equivalent with URIs of other datasets. However, there are not available connectivity measurements (and indexes) involving more than two datasets, that cover the whole content (e.g., entities, schema, triples) or “slices” (e.g., triples for a specific entity) of datasets, although they can be of primary importance for several real world tasks, such as Information Enrichment, Dataset Discovery and others. Generally, it is not an easy task to find the connections among the datasets, since there exists a big number of LOD datasets and the transitive and symmetric closure of equivalence relationships should be computed for not missing connections. For this reason, we introduce scalable methods and algorithms, (a) for performing the computation of transitive and symmetric closure for equivalence relationships (since they can produce more connections between the datasets); (b) for constructing dedicated global semantics-aware indexes that cover the whole content of datasets; and (c) for measuring the connectivity among two or more datasets. Finally, we evaluate the speedup of the proposed approach, while we report comparative results for over two billion triples.