A Self-Organizing Multi-Agent System for Stable Matchmaking with Incomplete Preferences in a Home Health Care Scenario

From Master Projects
Jump to: navigation, search


has title::a Self-Organizing Multi-Agent System for Stable Matchmaking with Incomplete Preferences in a Home Health Care Scenario
status: finished
Student name: student name::D. Ferro
number: student number::1283189
Dates
Start start date:=2006/02/01
End end date:=2010/02/01
Supervision
Supervisor: Jan Treur
Second reader: has second reader::Mark Hoogendoorn
Company: has company::Almende B.V.
Poster: has poster::Media:Media:Posternaam.pdf

Signature supervisor



..................................

Abstract

Title

a Self-Organizing Multi-Agent System for Stable Matchmaking with Incomplete Preferences in a Home Health Care Scenario


Abstract

For over more than three millennia, processes that match together people for business or social purposes have been an important aspect of daily life. In this thesis, we have designed and tested a multi-agent based matchmaking system for home health care. A case study project in the domain of home care service delivery, known as project \textit{Luister(english: Listen)}, is used to validate our approach. The system can handle matching requests for a particular care service made by individuals out of a group requesters, and assign each individual to an individual out of a group of providers. In this thesis, we are interested in the question how we can design such a matchmaking system, serving as a component in the ASK system, using a multi-agent system approach. We propose a framework for self-organizing matchmaking and we apply the Generic Agent Model to our case study, such that each agent represents a human actor : a patient, a volunteer or a professional. We have been able to implement a stable matchmaking interaction protocol, in which patient are the initiative taking party. To allow the agents to forecast unknown preferences, we use a number of different strategies. In addition, the peer expansion procedure allowed agents to expand their set of acquaintances. We have evaluated the performance of the system using a prototype with simulated data. Although, we have little basis for comparison, we can clearly state that agents are able to improve their mean feedback over time depending on the strategy used. The results show that the peer similarity based forecasting strategy and the peer \& service similarity based forecasting strategy work best. Although, a so-called popularity based forecasting strategy shows surprisingly high forecasting accuracy compared to the other strategies.

Final version Master thesis

thesis(pdf)