Innovative approaches to open-source data indexing

The collection and indexing of linked open data are crucial steps towards managing and exploiting the wealth of information on today’s Web. The aim of this project is to explore innovative ways for performing these functions, using Acropolis—the central part of BBC RES (Research & Education Space) project—as the main case study, thanks to a freshly started collaboration with BBC Wales (principal contact, Iain Tweedale ).

BBC RES indexes and organises the digital collections of museums, libraries and broadcasters to make their content more discoverable and usable. A central component is the aggregator, which examines the collected data, looking for instances where the same entity is described in more than one place (semantic alignment), and then aggregates and stores that information.

The project’s outcomes are twofold.

On the one hand, we aim at improving the aggregator by automating the semantic alignment, that currently is specified manually. The automatic version of such a semantic alignment will be defeasible, i.e., the system will go beyond logical inference in applying rules-of-thumb (traditionally probabilistic) to make inferences. These are usually employed when information is incomplete. Another source of incoherence for ontologies is inconsistency, i.e., when we get several facts that cannot be true simultaneously. Two well-known frameworks for addressing incoherence in knowledge bases are argumentation and belief revision. The former uses the generation of arguments and counter-arguments for or against a given conclusion. The latter provides operators for preserving consistency upon receipt of new information. This project will investigate the use of both approaches, giving rise to two alternative systems. After the implementation there will be an evaluation phase, consisting of experiments with end users.

On the other hand, we aim at exploiting the system’s user interface for receiving feedback via a conversational interface. The data created during this feedback will then provide the training data to be fed into a machine learning algorithm to improve the system over time.