Adaptive Ontologies Through Social Evolution (Doctoral Consortium)

The Semantic Web movement promotes the embedding of semantic content into Web resources. Unfortunately, the Web entities do not find enough motivation nor reward to do their part in the construction of a semantically richer Web environment. To enable a true semantic Web, each humanreadable document should provide a formal knowledge representation for its content. As hand-crafted knowledge acquisition seems infeasible when applied to an environment as vast and dynamic as the Web, efforts have been focused on automatic approaches to extract semantic content from textual documents. Advances in Natural Language Processing, Information Extraction and Ontology Learning provided tools that allow for the extraction and analysis of structured semantic knowledge given existing textual corpus. While these tools seem promising to enable a scenario where Semantic Web can be achieved, they need to be adapted to the scale and complexity of the Web. In this article, I propose that a possible solution lies in harvesting the power of emergent multi-agent societies to create an infrastructure capable of bootstrapping the adoption of Semantic Web technologies. I focus on how to create and adapt distributed on-line evolutionary algorithms to continuously design and improve social agents capable of semantic knowledge retrieval. I argue that currently existing automatic and semi-automatic structured knowledge acquisition techniques can be adapted to serve not only as building blocks for emerging and evolving knowledge extracting processes but also as one of the driving forces behind its adaptation. I then claim that learning extracting procedures can be done not only using annotated resources but also existing techniques creating dynamic fitness models that can be continuously updated to improve the system.