Semantic enrichment and exploration process on domain specific digital libraries

Semantic enrichment and exploration process on domain specific digital libraries

Fragkeskos, Kyriakos (TU Delft Electrical Engineering, Mathematics and Computer Science)

Lofi, Christoph (mentor)
Houben, Geert-Jan (graduation committee)
Venkatesha Prasad, Ranga Rao (graduation committee)

Delft University of Technology

2017-06-26

With the growing number of scientific publications, the concept
of navigating effectively and searching for domain specific information
is rather significant and highly important for the scientific
community [2]. For instance, to search by topics, research methods,
used datasets, or scientific objectives. Such deep meta-data increase
our perception for a given domain (i.e. Data Processing Pipelines) and
facilitate us to understand and visualize the evolution of research
topics and venues over time. Nevertheless, the extraction of such
deep meta-data from text-based documents is notorious challenging
and demanding due to the unstructured and ambiguous language
of the text in different publications.
The work in this paper has already contributed into two publication
attempts; with one published paper at ESWC conference,
and one accepted paper at the TPDL conference. Furthermore, the
work in this project extends the analysis of previous attempts by
adding more data from the domain of Data Processing Pipelines and
by including one additional domain for analysis (i.e. the domain
of Robotics). Moreover, this work provides justifications for all the
implementation decisions and proposes a refined version of the
online domain-aware semantic enrichment framework (SmartPub),
that automates the generation of deep meta-data by utilizing key
facets from the domains of Data Processing Pipelines and Robotics.
The goal is to generate structured meta-data (i.e. named entities or
phrases), from full-text scientific publications, with respect to a set
of domain aware facets (i.e. Objective, Methods, Dataset, Software,
and Results), and afterwards, to construct groups of facet-terms (i.e.
facet-topics) according to their semantic similarity for allowing
data exploration and navigation. Finally, the proposed framework
is evaluated both quantitatively and qualitatively on seventeen conference
series from the domains of Data Processing Pipelines and
Robotics.

Meta-Data Extraction
Digital Library
Facet Embedding
Natural Language Processing
Topic Modeling

http://resolver.tudelft.nl/uuid:4475a83f-d0e8-485a-ba74-5478f43c32aa

Student theses

master thesis

© 2017 Kyriakos Fragkeskos