John Davies, BT exact
The vision of semantic knowledge technology – SEKT for short - is to make knowledge truly accessible. Many organisations own large amounts of information in databases and intranets which is poorly structured, and not accessible according to the needs of a developing organisation. The Internet, accessible to anyone, constitutes the largest body of knowledge ever to have existed in the history of mankind. However, knowledge is only valuable if you can find it and make sense of it. That means it must be placed in a descriptive framework.
"Knowledge at your fingertips" means that knowledge can be retrieved from all relevant sources, and can be retrieved according to meaning, rather than keywords.
The Internet is full of unstructured text – comprehensible to people, but largely useless for computers. As a result, when you search for information with keyword, you often receive half a million hits, many irrelevant to your needs. The reason is that search machines can not use the contextual information which people have available to them all of the time.
SEKT is developing automated techniques for extracting meaning from the Web. By generating structured descriptions of Web pages, SEKT is making those pages machine-processable. So that in the future you will be able to get straight to what you want.
Searching intelligently When you use a search engine on the Internet or a corporate intranet, it has no understanding of the context of our search. For example, you might key in ‘European growth rate’, but the search engine has no idea what kind of ‘growth’ you are interested in: population growth, economic growth or some other measurement.
With SEKT you can establish a context for your search. Given semantically-annotated knowledge, SEKT software will exploit class hierarchies and attribute information to give a more intelligent response. For example, if you were looking for information about cars with certain properties (e.g. having the colour red), but nothing was available, SEKT could report the fact but also return information about red lorries, since it knows that cars and lorries are both types of vehicle. Similarly, SEKT could search for employees with specific attributes (e.g. over the age of 45) by exploiting extracted metadata.
Avoiding the information deluge How about if you wanted to search the Web for the speeches and writings of a famous politicians? You would need to use a textbased search, which will inevitably deluge you with redundant information about the politician. With SEKT you can specify explicitly that you want material written by the politician, not information about him or her. You will be able to specify that you want information written by a particular individual, about a particular topic, and containing a particular phrase or set of phrases.
Finding and sharing SEKT isn’t just about finding knowledge, it is also about sharing knowledge. Setting knowledge within a descriptive framework means it can be targeted at those who really need it. As a result, knowledge sharing between colleagues becomes a really valuable activity. Knowledge from different sources can also be integrated, and used to provide intelligent advice to support business and professional goals.
Part of the user’s environment To be really useful, SEKT needs to be an integral part of your working environment. SEKT is being designed to be seamlessly integrated into proprietary business software, and to support a variety of devices, such as PDAs and mobile phones.
The SEKT philosophy SEKT works with, not in place of, the user. No really satisfactory techniques exist for fully automating the process of describing documents. However, SEKT makes suggestions which can be accepted or overruled, giving the users the ability to specify how automated the process should be. The SEKT goal is to put the users in control, whilst freeing them from routine tasks to concentrate on value-creation.
The technology constituting SEKT
Knowledge discovery Knowledge discovery is concerned with techniques for automatic knowledge extraction rom data. It includes areas such as data and text mining, machine learning, statistics and parts of artificial intelligence. In SEKT these techniques will be mainly applied to textual data and in particular for (semi) automatic ontology generation, ontology evaluation, metadata extraction and ontology population.
Human language technology An understanding of the structure of human language, and of its syntax and semantics, will be used to complement the statistically-based knowledge discovery techniques. Taken together, knowledge discovery and human language technologies make for a powerful approach to ontology learning and metadata extraction.
Ontology management and reasoning Ontologies evolve over time, as the underlying knowledge evolves. This evolution needs to be managed. Moreover, not everyone will use the same ontology to describe a particular domain, so we need to be able to mediate between ontologies. For really intelligent search engines, we need to be able to reason using ontologies. In particular, robust minded reasoning will be needed to cope with the inconsistencies to be found on the global Web. SEKT will research state-of-the-art techniques to tackle all these requirements. These technologies represent key components of what is increasingly being known as the Semantic Web.
The ontologies used by SEKT will provide much richer descriptive frameworks than the strictly hierarchical taxonomies which the most advanced current commercial knowledge systems use.
Extending technical boundaries
Semantic knowledge technology is extending technical boundaries in a number of important ways: technically- for the SEKT research is not just about advances in basic technology. We will be investigating how the user best interacts with knowledge, not just at a computer terminal but also via a PDA or mobile phone.
SEKT will also be looking at how knowledge technology is best used in the organisation, and at how an ontological framework and the associated knowledge can best be presented to the user.
Open Powerpoint presentation (2645 kB) >>>>>
CLOSE WINDOW TO EXIT