Dynamic Matchmaking between Messages and Services in Multi-Agent Systems

1
Dynamic Matchmaking between Messages and Services
inMulti-Agent Systems

• Muhammed Al-Muhammed
• May 3, 2004

Support in part by NSF
2
Motivations

• Agents cooperate to achieve goals
• Cooperation needs communication
• Communication possible if agents
• share ontologies,
• speak the same language,
• pre-agree on a message format.

3
The Problem
Agents must
1- share ontologies,
2- speak the same language,
3- pre-agree on message format.

• Requiring these assumptions precludes
• agents from interoperating on the fly

The holy grail of semantic integration in
architectures is to allow two agents to
generate needed mappings between them on the fly
without a priori agreement and without them
having built-in knowledge of any common
ontology. Uschold 02
4
Solution
Agents must
1- share ontologies,
2- speak the same language,
3- pre-agree on message format.

• Eliminate all assumptions

- Dynamically capturing a messages semantics -
Matching a message with a service

5
Matchmaking System
(MMS)
MatchMaking System
Message-Service Matching
Message Handling
Response Handling
Global Domain Ontology
Services (Agent- Independent Representation)
Translation Repository
Service Analysis
Mapping
Translation
An Agent
Local Ontology
Services
6
Global Ontology Creation
(MMS)
MatchMaking System
Message-Service Matching
Message Handling
Response Handling
Concept Recognizers ProcessorType
(Processor)(TypeClass) ProcessorSpeed
(Processor)(Speed)(Processor)(Clock)(Speed) Uni
t of Measurement Recognizers ProcessorSpeed
Unit (GHzMHz)
Global Domain Ontology
Services (Agent- Independent Representation)
Translation Repository
Service Analysis
Mapping
Translation
An Agent
7
Local-Global Mappings (Initialization)
(MMS)
MatchMaking System
Message-Service Matching
Message Handling
Response Handling
Global Domain Ontology
Services (Agent- Independent Representation)
Translation Repository
Concepts (Local, Global) ------------------------
------------ (ProcessorClockSpeed,
ProcessorSpeed) (ProcessorClass,
ProcessorType) Units ProcessorSpeed GHz
Service Analysis
Mapping
Translation
An Agent double ProcessorClockSpeed
//GHz String ProcessorClass
8
Service Analysis (Initialization)
(MMS)
MatchMaking System
Message-Service Matching
Message Handling
Response Handling
Global Domain Ontology
Services (Agent- Independent Representation)
Translation Repository
Service Analysis
Mapping
Translation
An Agent public PcInfo getPcInfo (double
RAM) public int getPrice (String
ProcessorClass , double ProcessorClockSpeed )
//output Price public int getAmt(String
Processor ) //type definition class PcInfo
String ProcessorClockSpeed //GHz String
ProcessorClass
An Agent public PcInfo getPcInfo (double
RAM) public int getPrice (String
ProcessorClass , double ProcessorClockSpeed )
//output Price public int getAmt(String
Processor ) //type definition class PcInfo
String ProcessorClockSpeed //GHz String
ProcessorClass
9
Requests Rewriting (Initialization)
(MMS)
MatchMaking System
Message-Service Matching
Message Handling
Response Handling
Global Domain Ontology
Services (Agent- Independent Representation)
Translation Repository
Service Analysis
Mapping
Translation
An Agent String ProcessorClockSpeed
//GHz String ProcessorClass double Price
//US Price getPrice(ProcessorClockSpeed 2.6
GHz, ProcessorClass Pentium 4)
10
Matchmaking System (Operation)

MMS
MMS
Message-Service Matching
Message-Service Matching
Message Handling
Message Handling
Global Ontology
Global Ontology
Response Handling
Response Handling
Services (Agent- Independent Representation)
Services (Agent- Independent Representation)
Translation Repository
Translation Repository
Mapping
Mapping
Translation
Translation
Service Analysis
Service Analysis
Agent 2
Agent 1
?
11
Test Cases

• Real-World Test Cases
• Computer Shopping
• Book Shopping
• Meeting Scheduling
• Agents Coded w.r.t.
• Each web site (for shopping applications)
• Each individuals worksheet (for scheduling)

12
Agent Creation (Concepts Units)
13
Agent Creation (Services)
class PcInfo String ProcessorClass
String ProcessorSpeed //GHz
String
ReturnType? Name?( Type? InstalledMemory) Retur
nInformation?
14
Results (Computer Shopping, 9 Agents)
Tested Processes
Concept Recognition
Unit Recognition
Data Format Recognition
Concept Recognition
Total number of concepts in agents code 104
MMS-produced mapping pairs 94
Correct mapping pairs 91
Recall Recall ( of correctly recognized items) / (total of items that should have been recognized) 91/104 88
Precision Precision ( of correctly recognized items) / (total of recognized items) 91/94 97
15
Results (Computer Shopping)
Tested Processes
Concept Recognition
Unit Recognition
Data Format Recognition
Unit Recognition
Units Currencies US, GBP, EUR Number of instances in agents code 9
Units Capacity and speed GB, MB, GHz, MHz Number of instances in agents code 23
Total 32 32
MMS-recognized units 34 34
Correct units 32 32
Recall 32/32 100 32/32 100
Precision 32/34 94 32/34 94
16
Results (Computer Shopping)
Tested Processes
Concept Recognition
Unit Recognition
Data Format Recognition
Data Format Recognition
No data format of interest
17
Results (Book Shopping, 4 Agents)
Tested Processes
Concept Recognition
Unit Recognition
Data Format Recognition
Concept Recognition
Total number of concepts in agents code 27
MMS-produced mapping pairs 25
Correct mapping pairs 25
Recall 25/27 93
Precision 25/25 100
18
Results (Book Shopping)
Tested Processes
Concept Recognition
Unit Recognition
Data Format Recognition
Unit Recognition
Units Currencies US, EUR Number of instances in agents code 4
Units - -
Total 4 4
MMS-recognized units 4 4
Correct units 4 4
Recall 4/4 100 4/4 100
Precision 4/4 100 4/4 100
19
Results (Book Shopping)
Tested Processes
Concept Recognition
Unit Recognition
Data Format Recognition
Data Format Recognition
Data format Different date formats 3 Number of instances in agents code 4
Data format - -
Total 4 4
MMS-recognized data formats 4 4
Correct data format 4 4
Recall 4/4 100 4/4 100
Precision 4/4 100 4/4 100
20
Results (Meeting Scheduling, 4 Agents)
Tested Processes
Concept Recognition
Unit Recognition
Data Format Recognition
Concept Recognition
Total number of concepts in agents code 28
MMS-produced mapping pairs 22
Correct mapping pairs 22
Recall 22/28 79
Precision 22/22 100
21
Results (Meeting Scheduling)
Tested Processes
Concept Recognition
Unit Recognition
Data Format Recognition
Unit Recognition
No units of interest
22
Results (Meeting Scheduling)
Tested Processes
Concept Recognition
Unit Recognition
Data Format Recognition
Data Format Recognition
Data format Different date formats 4 Number of instances in agents code 4
Data format Different time formats 1 Number of instances in agents code 4
Total 8 8
MMS-recognized data formats 8 8
Correct data format 8 8
Recall 8/8 100 8/8 100
Precision 8/8 100 8/8 100
23
Contributions

• Built an MMS that lets agents communicate with no
  need to
• Share ontologies
• Use a common language
• Pre-agree on a message format
• Tested the MMS on three applications
• Concept mappings (90 accurate)
• Mappings for units and data formats (98
  accurate)

24
Future Work

• Generalize the recognizers and adding some
  reasoning rules
• Extend the matchmaking capability to cover
  partial matching
• Handle all types of knowledge sharing among
  agents.