New agent produces smarter text searches
A major problem today is not lack of information but the overwhelming mass of it. Many people feel that the pressure from this leads to information stress – difficulties in finding the right info, on the right occasion, and sensibly and intelligibly structured. Search engines on the Web dig out a plethora of data, much of it not even what we are looking for.
Research teams at KTH and Stockholm University are therefore planning to devise new and smarter tools to remedy this situation. Project KEA, ”KunskapsExtraktionsAgent”, or Knowledge Extraction Agent, has just received SEK 2.2 million from VINNOVA for a three-year study. This will directly involve two practical applications: the huge medical case-book systems at Stockholm County Council, and Agent25, the media monitoring company, with its large info bank.
– This is a quite exciting prospect – applying basic research efforts in language technology directly on two hot information areas, such as medicine and business intelligence, says project leader Dr. Hercules Dalianis at the Dept. of Computer & Systems Science (a joint KTH/SU department). Also, the specific communication problems which the council and Agent25 are having here will be most interesting to study.
KEA is meant to enable its users to find connections between different texts, including pertinent information in other languages than the search idiom used, also to create an automatic summary from one or more pertinent texts of a copyright nature, cf. below.
The Stockholm County Council wants an automatic system that may structure the huge number of case-book texts stored in the internal data systems of their various hospitals. In addition, the council´s wish-list includes that of tracking visible as well as invisible connections between different patient factors, for instance illnesses, medication, sex, age, profession, lifestyle, etc.
Agent 25 does have a need to make automatic summaries of the 40,000 texts they are receiving every day, in some ten different languages. One of the reasons for this is the current copyright legislation – the company may use original texts only to a very limited extent.