KESO (Knowledge Extraction for Statistical Offices)
KESO (Knowledge Extraction for Statistical Offices) is an ESPRIT III funded research and development project aiming at developing a prototype knowledge extraction system to suit the particular needs of statistical offices.
The project started in the beginning of 1996 and the final product is due in the beginning of year 1999.
The project involves two types of partners in four European countries:
- Developers from research institutes doing research on data mining, and a software house to integrate the results of the developer parties to a consistent system.
The developer parties in the KESO project are:
- End users from statistical offices to describe their specific needs to the developers and to give feedback after gaining experience using the preliminary releases of the KESO system.
The statistical offices in the KESO project are:
Keso will be developed as a series of prototypes embedding an increasing
set of search strategies, result refining and user interaction.
KESO v.1 was released in February 1997, and the second release in the
spring of 1998. The final release is due in March 1999.
Each release will be handed to the end users who will gain hands-on
experience and produce useful feedback to improve the further releases.
Structure of KESO
KESO consists of the following parts:
- Mining Conductor
to guide the process by
- instantiating search modules with tasks (parallel execution)
- deciding what to do when some search task is finished (successfully or not)
- communicating with the user interface
- Search Space Manager
to maintain all information in the search space
- A set of Search Modules
where each search module performs a search task (of given type) by
- selecting a hypothesis to be evaluated
- asking the corresponding description generator to generate new hypotheses
- A set of Description Generators
to generate new hypotheses from previous ones
- Quality Computer
to evaluate the hypothesis against the actual database
- Mining Server
to query the actual database (stored in the Monet database system)
- Human-Computer Interface
to allow the user to
- request for a search task
- refine his/her request according to the results obtained
University of Helsinki in KESO
Presently, the UH team consists of the following researchers:
Our main tasks in developing KESO are involved with
- Association Rules
the description generator and quality
function for association rules are implemented in KESO
- Result Refining
- how to describe (and limit) the request?
- how to present the results?
- Time Series
a similarity measure for time series will be
implemented in the next version of KESO
- Discovery of Hierarchies
- Performance of the KESO System
January 23, 1999
verkamo@cs.Helsinki.FI