In this paper, an agent is defined as a triple (S, R_S, L_S), where S is a multi-hierarchical decision system, R_S is a set of rules extracted from S defining values of its decision attributes, and L_S is a language which the agent can use to communicate with other agents. L_S is built from values of decision attributes in S which are treated as agent's external attributes. Classification attributes in S are treated as agent's internal attributes which for all agents are the same. If objects stored in two decision systems representing different agents are the same, then their descriptions in terms of internal attributes are the same as well. Agents can learn from each other definitions of their external attributes. If these definitions differ or are contradictory then agents may try to propose a new definition which is more acceptable to both of them. Standard semantics and agent centered semantics are introduced and used to describe the strategy of reaching consensus among agents. Expressions in the language L_S are called analytical questions. Music information retrieval is taken as an application domain.
展开▼
