research pages

Computer-mediated communication of structured knowledge

These pages contain a high level description of work that has been performed and obtained results. Some of this work may be discontinued. The detailed presentation of these results is in the cited papers.

Goal and motivation

We assume that expressing formalised knowledge on a computer is useful, not especially for the need of the computer, but for communication. In future information systems, formalised knowledge will be massively exchanged. Exmo's goal is the development of theoretical and software tools for enabling interoperability in formalised knowledge exchange. Exmo contributes to an emerging field called the semantic web which blends the communication capabilities of the web with knowledge representation.

There is no reason why knowledge should be expressed in a single format or by reference to a single vocabulary (or ontology). In order to interoperate, these representations will have to be matched and transformed. Moreover, in the communication process computers can add value to their memory and medium role by formatting, filtering, classifying, consistency checking or generalising knowledge.

We currently build on our experience of alignments as representing the relationships between two ontologies on the semantic web.

Ontology alignments express correspondences between entities in two ontologies. They allow maximising sharing on the semantic web: various algorithms can produce alignments and various uses can be made of these alignments. Such alignments can be used for generating knowledge transformations (or any other kind of mediators) that will be used for interoperating. In order to guarantee properties of these transformations, we consider the properties of alignments and generate transformations preserving them.

Our current roadmap focusses on the design of an alignment infrastructure and on the investigation of alignment properties (and especially semantic properties) when they are used for reconciling ontologies.

On a longer term, we want to explore « semiotic » properties, i.e., properties which concern the interpretation of the communicated representation by a human user. This goal should require an analysis of the extra-semantic rules that govern the choice of subsets of models.

Anticipated applications are in transformation system engineering (in which the information system is seen as a transformation flow) and the semantic web infrastructure.

Achievements

We summarise six recent outcome of our work that we think particularly important. There are many other results, accessible through the table of content.

Semantics of networks of ontologies

We have introduced a semantics for networks ontologies related by alignments. This semantics considers alignments as constraining the meaning of ontologies so that each ontology is made more precise and entails more knowledge. This semantics is used for defining other operators on the networks such as consequence computation, alignment composition or network revision.

Revision in networks of ontologies

Given a semantics, networks of ontologies can be found as inconsistent when adding new formulas or correpondences. We have defined belief revision for networks of ontologies in order to restore consistency in such networks. We showed that local revision of ontologies or alignements was not able to implement revision in case of global inconsistency and provided partial meet revison operators for networks of ontologies.

Link keys for data interlinking

We introduced the notion of link keys as a generalisation of database keys which identify individuals across different data sets. We developed link key extraction techniques considering a limited amount candidate link keys. These candidate have been characterised as formal concepts. We provided measures to select the most efficient candidates with or without sample links. These provided accurate results in experiments.

Querying RDF graphs with paths

We have extended the standard SPARQL query language for dealing with paths of unknown length and provided complete algorithms for answering queries. This extension, PSPARQL, has been shown sufficient for answering SPARQL queries modulo RDF schemas. This language has been further extended for supporting constraints. This is sufficient to answer CPSPARQL queries modulo RDF Schema.

Ontology distance and matchers

We design metrics to be used for comparing ontologies and entities which are available in the OntoSim library. We develop ontology matchers based on these measures (Aroma, OLA). Ontology metrics can be used in other applications such as finding peer affinity in semantic social networks.

Alignment API and server

We develop and maintain an API and implementation of alignment structure which is used by more than 40 teams around the world. It allows for piping matching algorithms, manipulating alignments (trimming and hardening), generating processing output and comparing alignments. This API is also embedded in an Alignment server providing support for storing, finding, and sharing alignments over various communication means (web service, http, agent languages). The Alignment API is used in yearly matching evaluations.

Content

After setting the applicative context:

Our programme is presented around major axes:

Content and knowledge representation is the kind of languages we manipulate,
Ontology alignments for representing correspondences between models:
Ontology matching aims at finding alignments that can be used as transformations:
Data interlinking aims at finding links between linked data sets:
Transformations and properties is the general framework that we promote and apply to various contexts:
Semantic interoperability focuses on semantic properties of transformations:
Distributed and dynamic aspects of matching and exploiting alignments:
Infrastructure presents the steps toward a safe tranformation infrastructure:
Computational semiology, on a longer term, considers the opportunity to transmit more context during the communication.