Semantic Web Technologies

     MultiAgent Systems
     Electronic Commerce

     MAS Management

     Information Gathering
Multi-Agent Learning
MAS Interoperability


Project Grants
Site Map

In order for heterogeneous agents to coordinate effectively across distributed networks of information, they must be able to communicate with each other using a common language. To address the problem of agent interoperability, we are developing an agent capability description language called LARKS ( L anguage for A dvertisement and R equest for K nowledge S haring). This common language is used by middle or matchmaking agents to pair service-requesting agents with service-providing agents that meet the requesting agents' requirements.

When a service-providing agent registers a description of its capabilities with a middle agent, it is stored as an "advertisement" and added to the middle agent's database. Thus, when an agent inputs a request for services, the middle agent searches its database of advertisements for a service-providing agent that can fill such a request. Requests are filled when the provider's advertisement is sufficiently similar to the description of the requested service.

LARKS is expressive, easy to use, and capable of supporting inferences. It also incorporates application domain knowledge in agent advertisements and requests. Domain-specific knowledge is specified as local ontologies in the concept language ITL.

An advertisement or request in LARKS comprises the following sections:

Context Keywords identifying the domain
TypeDefinitions User-defined data types
InputDeclarations and OutputDeclarations Input and output parameter declarations
InConstraints and OutConstraints Logical constraints on input and output parameters
ConceptDefinitions Local ontological description of words used in the advertisement
TextDefinition A text description of the agent's capabilities

The LARKS matchmaking process employs techniques from information retrieval, AI, and software engineering to compute the syntactical and semantic similarity among agent capability descriptions. The matching engine of the matchmaker agent contains five different filters for context matching, word frequency profile comparison, similarity matching, signature matching, and constraint matching. The user configures these filters to achieve the desired tradeoff between performance and matching quality.

We have implemented a matchmaking agent and a visually expressive user interface that traces the path of a service request through the matchmaker's five filters.

Robotics Institute Project Page

Larks Applications
Larks is used in the following applications:

For more information on the LARKS project, see the following publications:


Copyright 2006 - 2012 © Advanced Agent-Robotics Technology Lab - The Robotics Institute - Carnegie Mellon University

Internal Site (Restricted Access).