Master Thesis Defense by Ms Medina-Petrina Andresel
Ms Medina-Petrina Andresel defended her master thesis on 'A Compilation Technique for Interactive Ontology-mediated Data Exploration'
Ms Medina-Petrina Andresel defended her master thesis on 'A Compilation Technique for Interactive Ontology-mediated Data Exploration' at TUW on 4 March 2016.
Abstract: Ontologies have been extensively advocated in the last decade as a
means to enable a shared understanding of resources between different
users and applications. One of their championed applications is Ontology-
based Data Access (OBDA), where ontologies mediate access to data
sources, providing a high-level conceptual view of the data and making
background knowledge available for reasoning at query time. Much re-
search efforts in the last years have been dedicated to the core problem
in OBDA: answering a query while leveraging the knowledge given by an
ontology, sometimes called ontology mediated query answering (OMQA).
However, most work focuses on providing algorithms for answering queries
one-at-a-time. In this thesis, we make a first step towards interactive
answering queries, that are syntactically related, and exploring targeted
fragment of data, in the ontology-mediated setting. We introduce a technique to
construct an offline compilation that allows, in online phase,
to effciently answer (without accessing the original data source) different
variations of queries that fall between two query bounds (given as
input to capture user's information needs from above and below). We
consider ontologies formalized using DL-Lite - the most popular ontology
language, and we focus on conjunctive queries (CQs) that are tree-shaped
and have one answer variable - relevant for practical purposes by allowing to capture
objects that can have complex combinations of properties.
The compilation represents the underpinning for constructing relevant
query modiffications, which help the user analyze the data by reducing
or increasing the number of answers in a minimal way, or identifying all
most specific common properties that the answers of a given query share.
The experiments carried out with our prototype implementation reveal an
overall good performance of the query-answering procedure and, most importantly,
our solution seems to be a promising first step towards
flexible ontology-mediated data exploration.