RDF (Resource Description Framework ) is the Semantic Web core model which is used to represent data thanks to the definition of triples. RDF data editing is often a manual task which may seem tedious for the concerned users, and which includes an important risk of errors. A Web editing tool along with a suggestion system have been created to assist the editing of the Henri Poincaré correspondence corpus. The tool uses RDFS (RDF Schema) entailment and case-based reasoning to provide relevant suggestions when editing a resource. Four versions are proposed and are compared to each other through a double evaluation. While this system has been created for and evaluated with the Henri Poincaré correspondence corpus, the proposed methods could be applied to other corpora.
Manual rdf data editing is the task which consist in creating triples through a
form-based interface or directly by writing them to a file using an adapted rdf
serialization format (e.g. rdf/xml, Turtle). The Henri Poincaré correspondence
corpus is centered around 2100 sent and received letters which constitute
scientific, administrative and private exchanges. The rdf model has been chosen to
represent corpus data, the rdfs language to represent ontology knowledge and
the sparql language to query data. The manual editing of corpus data has been
identified as a tedious task which requires the constant attention of the user to
avoid different kinds of mistakes. A Web editing tool along with a suggestion
system have been developed to assist users in this editing process. This system
has already been presented in the context of the International Semantic Web
Conference [
Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
Presentation of the system
The goal of the suggestion system is to provide a ranked list of potential values
associated to a field (i.e. subject, predicate or object) when editing a triple. Four
versions of the suggestion system have been designed. The basic version orders
the suggestions using the alphabetical order. The deductive version benefits from
the use of the ontology knowledge when ranking the potential values. In
particular, it takes advantage of the knowledge about the domain and range of the
ontology properties. The case-based version uses case-based reasoning to exploit
knowledge about similar edited resources. As an example, when one is editing a
letter and wants to set its recipient, it may be relevant to search for letters having
the same topics discussed, quoting the same individuals or institutions, or
written within a relatively close temporal interval. A system based on sparql query
transformation rules is reused to find resources similar to the resource which is
currently being edited. This system, named sqtrl (sparql Query
Transformation Rule Language), has proven useful in various contexts [
Associated with the suggestion system, a Web user interface has been developed to visualize and to update rdf databases. This interface proposes an autocomplete mechanism that uses the suggestion system for providing values when editing a resource. A demonstration video of this tool in use for the Henri Poincaré correspondence corpus is available online.
Two evaluations have been conducted for this system. The first evaluation is human-based through a user who edits around 30 letters of the Henri Poincaré correspondence corpus using the Web editor. At the end of the editing, he/she is invited to complete a survey to give his/her feedback about the four suggestion system versions. A second evaluation is managed through a dedicated program which provides objective measures by simulating editing situations based on already edited content of the corpus. Both evaluations results suggest that the version combining rdfs entailment and case-based reasoning is the most efficient one and could significantly assist users in the editing process.
Acknowledgement. This work is supported partly by the French PIA project “Lorraine Université d’Excellence”, reference ANR-15-IDEX-04-LUE.