LINNEO+ is a knowledge acquisition tool and it works incrementally with an Unsupervised learning strategy that accepts a stream of observations and discovers a classification scheme over the data stream. As a control strategy it retains only the best hypotheses that are consistent with the observations given a similarity criterion.
LINNEO+ uses information about the domain elements to induce classes and takes advantage of domain theory (a priori knowledge) which represents the expert's current state of knowledge (which may be incomplete). This theory constrains the possible outcomes of the classification process semantically biasing the results.
The human expert abstracts a collection of observations and a set of attributes that he thinks of as relevant . At the same time, the expert is also able to express what he already knows about the domain. This knowledge (that will be called Domain Theory (DT)) is represented as set of rules. Starting with this knowledge and data, LINNEO uses induction over the observations to generate a classification (see [Béjar95]).
The human expert abstracts a collection of observations and a set of attributes that he thinks of as relevant . At the same time, the expert is also able to express what he already knows about the domain. This knowledge (that will be called Domain Theory (DT)) is represented as set of rules. Starting with this knowledge and data, LINNEO uses induction over the observations to generate a classification (see [Béjar95]).
The LINNEO+ tool has been fully developed inside our group.
References
- [Béjar95]
- J. Béjar, Adquisición de conocimiento en dominios poco estructurados PhD thesis, Departament de Llenguatges i Sistemes Informàtics. Facultat d'Informatica de Barcelona. Universitat Politecnica de Catalunya, 1995.
- [Béjar94]
- J. Béjar, U. Cortés,and M. Domingo, 'Using domain theory to bias classification processes in ill-domains', in proceedings of the IV congreso Iberoamericano de Inteligencia Artificial (IBERAMIA '94), pp. 187-197, 1994.
The KEMLg members involved:
|
Share: