Databases Reference
In-Depth Information
The second layer of FOX is the
controller
, which coordinates the access to the mod-
ules that carry out the language processing. The controller is aware of each of the
modules in its backend and carries out the initialisation of these modules once FOX
is started. Furthermore, it collects the results from the backend modules and invokes
the results of a training instance to merge the results of these tools.
The final layer of FOX is the
tool layer
, wherein all NLP tools and services integrated
in FOX can be found. It is important to notice that the tools per se are not trained during
the learning phase of FOX. Rather, we learn of the models already loaded in the tools
to allow for the best prediction of named entities in a given domain.
The ensemble learning implemented by FOX was evaluated in the task of NER by
integrating three NER tools (Stanford NER, Illinois NER and a commercial tool) and
shown to lead to an improvement of more than 13% in F-Score (see Figure 7) when
combining three tools, therewith even outperforming commercial systems.
100,00
90,00
80,00
70,00
60,00
Stanford NER
Illinois NER
Commecial Tool
FOX
50,00
40,00
30,00
20,00
10,00
0,00
Precision
Recall
F-Score
Fig. 7.
Comparison of precision, recall and F-score of the best runs of FOX and its components
on NER
3.2
From Structured Sources
Structured knowledge, e.g. relational databases and XML, is the backbone of many
(web) applications. Extracting or converting this knowledge to RDF is a long-standing
research goal in the Semantic Web community. A conversion to RDF allows to integrate
the data with other sources and perform queries over it. In this lecture, we focus on the
conversion of relational databases to RDF (see Figure 8). In the first part, we summarize
material from a recent relational database to RDF (RDB2RDF) project report. After
that, we describe the mapping language R2RML, which is a language for expressing
database to RDF conversion mappings. While we focus on relational date, we also want
to note that extraction from CSV files is also highly important as illustrated in use cases
in the financial [96] and health sector [165,166].
Triplify and RDB2RDF Survey Report.
The table displayed in Figure 9 is taken from
the Triplify WWW paper [8]. The survey report [135] furthermore contained a chart(see
Figure 10) showing the reference framework for classifying the approaches and an ex-
tensive table classifying the approaches (see Figure 11). Another recent survey is [144].
