PrOQAW: Probabilistic Ontological Query Answering on the Web

The next revolution in Web search as one of the key technologies of the Web has just started with the incorporation of ideas from the Semantic Web, aiming at transforming current Web search into some form of semantic search and query answering on the Web, by adding meaning to Web contents and queries in the form of an underlying ontology. This also allows for more complex queries, and for evaluating queries by combining knowledge that is distributed over many Web pages, i.e., by reasoning over the Web.

Realizing such semantic search and query answering on the Web by adding ontological meaning to the current Web conceptually means annotating Web pages and their contents relative to that ontology, i.e., relating Web pages and their contents to and thus also via that ontology. From a practical perspective, one of the most promising ways of realizing this is to perform data extraction from the current Web relative to the underlying ontology, store the extracted data in a knowledge base, and realize semantic search and query answering on this knowledge base. There are recently many strong research activities in this direction.

A major unsolved problem in the above context is the principled handling of uncertainty: In addition to natural uncertainty as an inherent part of Web data, one also has to deal with uncertainty resulting from automatically processing Web data. The former also includes uncertainty due to incompleteness and inconsistency in the case of missing and over-specified information, respectively. The latter includes uncertainty due to, e.g., the automatic annotation of Web pages and their contents, the automatic extraction of knowledge from the Web, matching between different related ontologies, and the integration of distributed Web data sources.

The central goal of the proposed research is to develop a family of probabilistic data models for knowledge bases extracted from the Web relative to an underlying ontology, along with scalable query answering algorithms, which may serve as the backbone for next-generation technologies for semantic search and query answering on the Web. We believe that such probabilistic data models and query answering algorithms can be developed by integrating ontology languages, database technologies, and formalisms for managing probabilistic uncertainty in the context of the Web. The objectives include developing probabilistic data models, developing algorithms for ranking and query answering, identifying useful scalable fragments, and practically evaluating our results.

Towards next-generation technologies for semantic search and query answering on the Web, the project's central goal is to develop a family of probabilistic data models, along with scalable query answering algorithms, for knowledge bases that are extracted from the Web relative to an underlying ontology. We believe that these data models and query answering algorithms will exert a major influence on the theory and practice of data extraction from the Web and of semantic search and query answering on the Web. Thus, short-term non-academic beneficiaries will include industrial researchers in these fields. So, unsurprisingly, companies working on the former, such as Lixto, and those working on the latter, such as Yahoo!, Google, and Microsoft, have reported a strong interest in the project (Lixto, Yahoo!, and Google by letters of support, and the director of Microsoft's FUSE (Future Social Experiences) Lab in Cambridge in personal communication with the PI).

In longer terms, our research will also lay the foundation for a new generation of information systems that will allow for dealing with incomplete, uncertain, inconsistent, semi-structured, overlapping, and semantically related data. Sources of such data are spreading at a phenomenal rate, most notably in the context of the Web. Thus, in the mid term, non-academic beneficiaries will include industrial researchers who are trying to develop information systems aimed at dealing with this kind of data, as well as industrial researchers and developers who would like to exploit such data in applications. For example, as for the latter, Alcatel-Lucent have uncertain data from community-curated content and sensor readings, which they would like to exploit (see letters of support).

In the long term, non-academic beneficiaries could include anyone using or depending on the Web or on any other information system, which essentially includes every individual and every business and organisation. In particular, any significant advances in semantic search and query answering on the Web may influence all our lives in a similar, revolutionary way as the invention of the technologies behind Google's Web search. For example, one long-term vision of Web search is the one of an intelligent query answering interface in spoken natural language, similar to a human being; such an interface would make the Web accessible to a much larger class of people than nowadays (e.g., by spoken language via mobile phones) and would offer highly exciting new economic and social opportunities, contributing to UK's health, wealth, and culture. For example, a vertical such interface for a specific domain could offer completely new educational opportunities, reaching a much larger class of people at much lower costs than traditional ones. The proposed research will lay the foundations in this direction.

Our contacts and collaborations with industry (see above) will help us to ensure that our work will also have an immediate impact on and benefits for companies and organisations that develop and use Web information systems. For example, the Web data extraction applications developed by Lixto, and the semantic search initiatives for the Web by Yahoo!, Google, and Microsoft all would directly benefit from the results of this project. As non-academic dissemination and engagement activities, in addition to continuing our collaborations with industry, we will also explore the possible commercialisation of our results (e.g., via these collaborations).