ppt

i-footman: A Knowledge-Based
Framework for Football Managers
Vassilis Papataxiarhis, V.Tsetsos, I.Karali, P.Stamatopoulos, and
S.Hadjiefthymiades
[email protected]
Department of Informatics and Telecommunications
University of Athens – Greece
RuleApps-2009, 21 Sep. 2009, Cottbus
Outline

Introduction

Functionality and Provided Services

Application Models and Rules

Implementation

Simulation Results

Conclusions
Introduction

What is i-footman?
◦ A decision support system for football managers
◦ Based on Semantic Web technologies

Main Idea
◦ Provide effective tactical guidelines to face an opponent

Restrictions
◦ Empirical/Subjective Knowledge about football
◦ Lack of statistics and ergometric results
◦ No relevant approach (academic or industrial)

Goals
◦ Model the basic knowledge of the domain
◦ Extensibility (in terms of quality and provided services)
Knowledge Elicitation

Methodology
◦ Interview 2 domain experts (i.e. football
managers)
◦ Questionnaires

Knowledge acquisition about:
◦ the application domain of football
◦ the desired services
◦ the key features of football players and teams
◦ the tactical guidelines that should be
supported by the system

Goal: Incorporate the derived knowledge to
the rules and application models
i-footman Architecture
DL-Reasoner
i-footman
reuses
Football Players
Ontology
reuses
user
Football Teams
Ontology
reuses
Rules
Formation
Identification
Player Selection
Tactical Instructions
Rule
Engine
Functionality
Teams Data
Formation and
Player Selection
Rules
Football
Players
Ontology
Formation
DL-Reasoning
Rules
Execution
Composition
Strengths/
Weaknesses
Instructions
Football
Teams
Ontology
Players Data
Identification and
Tactical
Instructions Rules
Ontological Models
(1/2)

Expressed in OWL-DL and provide a common vocabulary

Football Players Ontology (FPO)
◦ Some metrics: 71 concepts, 43 object prop., 3 datatype prop., each
player instance is described by 22 concept inst. and 9 property inst.
◦ It models:
 Position of players
 Technical and physical capabilities
 Types of players
E.g., fpo:CreativeMiddlefielder≡ (fpo:hasPassing.GoodAbility ⊔
fpo:hasPassing.VeryGoodAbility) ⊓
fpo:playsInPosition.Middlefielder

Football Teams Ontology (FTO)
◦ It models main features and types of teams
Ontological Models

Simplified version of FPO

Key concepts
◦ Player, Position, PlayerFeature

FTO imports FPO
◦ classifies teams according to
the features of its players
◦ models tactical instructions
allowing the execution of
rules
(2/2)
Rules

Expressed in terms of SWRL
◦ Motivation: integration of rules and ontologies in the same logical
language





Exploit the vocabulary of FPO and FTO
Define more complex concepts and relationships
Constitute the main part of the knowledge acquired by
interviewing the experts
Extensible set of rules
Four main categories of rules for the:
◦
◦
◦
◦
identification of team weaknesses/advantages
selection of an appropriate tactical formation
player selection
recommendation of appropriate tactical instructions
Rules Examples

Identification Rule
◦ fto:hasStartingPlayer (?t1,?p1) ∧ fto:hasStartingPlayer (?t1,?p2) ∧
fpo:QuickOffensivePlayer (?p1) ∧ fpo:QuickOffensivePlayer (?p2) →
fto:dangerousAtCounterAttack (?t1,true).

Formation Rule
◦ fto:myTeamPlaysAgainst(?t1,?t2) ∧ fto:TeamWith3CentralDefenders(?t2)
∧ fto:TeamWith3CentralPlayers(?t2) ∧ fto:TeamWithSideMFs(?t2) ∧
fto:TeamWith2Attackers(?t2) → fto:playsWith3CentralDefenders(?t1,
true).

Player Selection Rule
◦ fto:myTeamPlaysAgainst(?t1,?t2) ∧ fpo:playsWith1Striker(?t1) ∧
fpo:GoodStriker (?p1) ∧ fpo:isMemberOf(?p1,?t1) →
fpo:isSuggestedTo(?p1,?t1).

Tactical Instruction Rule
◦ fto:myTeamPlaysAgainst(?t1,?t2) ∧ fto:TeamWithNoBacks(?t2) ∧
fto:TeamWithWingers(?t1) → fto:shouldAttackFromTheWings(?t1, true).
Implementation Details

Web Ontology Language (OWL-DL)

Semantic Web Rule Language (SWRL)

Pellet Reasoner (v. 1.5.1)

Jess Rule Engine

Protégé SWRL Jess Tab

Protégé OWL API

SPARQL

Jena2 inference module – Jena API

Apache Tomcat
Evaluation


(1/2)
Simulation of football matches in 2 platforms with and without the
intervention of i-footman
2 Scenarios
◦ Teams with similar ratings
◦ i-footman controls a weaker team

40 games in each platform (80 games in total)

Scenario 1
Barcelona FC vs. Real Madrid FC (goals)
Barcelona FC vs. Real Madrid FC (match results)
18
16
14
12
10
8
6
4
2
0
17
14
15
70
16
CPU
CPU
61
60
i-footman
i-footman
51
50
45
10
8
36
40
30
20
10
0
Wins
Draws
Losses
Goals +
Goals -
Evaluation
(2/2)
Scenario 2

Olympiacos SFP vs. Real Madrid FC (goals)
Olympiacos SFP vs. Real Madrid FC (match results)
100
30
CPU
26
i-footman
25
i-footman
80
70
20
18
17
62
60
50
15
11
40
10
31
30
5
5
CPU
87
90
23
20
3
10
0
0
Wins
Draws
Losses
Goals +
Goals -

No significant improvement when controlling a better team

Performance Evaluation
20%
38%
Average Response Time =
7740ms
Classification
Instance Checking
Rule Execution
42%
Conclusion


Contributions
◦ A knowledge-based system based on SW technologies
◦ An extensible framework for football managers
◦ FPO, FTO ontologies
Open Issues
Inferred
◦ Integrated reasoning module for
handling rules and ontologies
seamlessly
◦ Real data are not available

Knowledge
Rules
Execution
Ontological
Reasoning
Inferred
Knowledge
Future Work
◦ Automated ontology creation by statistics and ergometric data
◦ Learning rules by historical data stemmed from simulations
without the intervention of i-footman
◦ Adoption of fuzzy approaches to deal with uncertainty
Thank you!
http://www.di.uoa.gr/~vpap/i-footman