A Curtin University of Technology researcher has used artificial intelligence to develop new software that learns as it is used, an advance that could make online mapping and GPS more user-friendly.
Mr Matthew Hutchinson developed the prototype software, IntelliGeoLocator, as part of his PhD on making geocoding of street addresses more accurate and easier to use.
A geocoding process takes a reference to a physical street address and determines its’ location on the earth’s surface.
“The purpose of my research has been to change the paradigm of geocode processing, creating software that learns from its interactions with the user,” he said.
“The rationale behind this change in focus is that there are property and street addresses for which current geocoding software cannot resolve their locations.
“After a period of great progress, mapping programs are coming close to reaching their full potential using current geocoding techniques.”
It is currently estimated that between 5-10 per cent of addresses will require new techniques to accurately geocode.
As recently as 2003, five per cent of addresses in rural areas were inaccurate by about three kilometres or more, and five per cent of all suburban addresses could not be geocoded within 421 meters.
Although this has been improved significantly, many areas are still not accurate enough for purposes such as responding to emergency calls.
“Problems can be caused when there are two streets with the same name in the region, or a street number on the wrong side of the road,” Mr Hutchinson said.
“Other complications, such as abbreviated addresses or addresses in private estates can also arise, when the person describing the address uses the estate instead of the suburb. One way around this is to use semantics.”
Using semantics, Mr Hutchinson believes that geocoding programs will be better able to associate what the user thinks is correct — whether or not it is — with the reality of addresses in existence.
Semantics involves the differences in meaning that occur when we use natural, everyday language for modelling the real world in digital databases. Natural language can be subjective and people can also use different vocabularies and descriptions for address locations.
Further development in learning and semantics in geocoding would allow intelligent geocoders to infer new knowledge based on the information it already possesses.
“This effectively means that the software will be able to learn,” Mr Hutchinson said.
“The result would mean that if someone inputted the word ‘park’, the geocoder could also try the words ‘reserve’ and ‘oval’ to see if they should be used instead.”
These forms of improvements mean that geocoding software will become more personalised and contextualised to specific types of users and geographic locations.
“It is envisioned that with further work, geocoders will increasingly use intelligent information about the user or about the location to obtain an accurate geocode result” Mr Hutchison said.
“Who the user is, where they are and what they use the results for, all have implications for the required accuracy of a geocode.”
The framework used in Mr Hutchinson’s prototype would be usable in online mapping products and GPS devices, making them more user-friendly.
“The prototype has shown the ability to use intelligence to identify locations of problem addresses,” he said.
“Given the increasing importance of geographic information, the use of intelligence in geocoding will really enhance applications that rely on location.”
Mr Matthew Hutchinson completed his PhD through Curtin’s Department of Spatial Sciences, a part of the Western Australian School of Mines.
