=Paper=
{{Paper
|id=Vol-511/paper-4
|storemode=property
|title=i-footman: A Knowledge-Based Framework for Football Managers
|pdfUrl=https://ceur-ws.org/Vol-511/paper4.pdf
|volume=Vol-511
}}
==i-footman: A Knowledge-Based Framework for Football Managers==
https://ceur-ws.org/Vol-511/paper4.pdf
i-footman: A Knowledge-Based Framework for
Football Managers
Vassilis Papataxiarhis, Vassileios Tsetsos, Isambo Karali, Panagiotis
Stamatopoulos, Stathes Hadjiefthymiades
Department of Informatics and Telecommunications, National and Kapodistrian
University of Athens, Panepistimiopolis, Ilissia, GR-15784, Athens, Greece,
{vpap, b.tsetsos, izambo, takis, shadj}@di.uoa.gr
Abstract. i-footman constitutes a knowledge-based framework aiming to
provide assistive services to football managers. More precisely, the system
accommodates a number of managing processes (e.g., selection of team
composition) in the context of football by adopting a declarative approach. It
also takes advantage of Semantic Web technologies in order to represent and
manage the required application models. i-footman is based on a flexible
architecture that facilitates possible extensions in the functionality and the
quality of the provided services. The paper presents the overall architecture as
well as certain implementation details of the system.
Keywords: knowledge-based system, reasoning methods, ontology, rules
1 Introduction
This work presents an extensible knowledge-based framework aiming to provide team
management services in the context of football. The system allows the development
of services that can be used by football managers. Specifically, i-footman (intelligent
football manager) supports some pre- and in-game decisions needed to be taken by the
user. The basic functionality of i-footman consists of proposing a “good” tactical
formation, an effective composition of players and certain tactical instructions.
Football does not constitute a scientific field or a domain with explicitly expressed
and commonly accepted knowledge. This is due to the fact that knowledge about
football does not stem only from expertise but also from experience. Moreover, there
are times when domain experts (i.e. football managers) act differently to each other in
order to face a certain situation (e.g., suspension of a player) making the knowledge
about football subjective. Hence, i-footman does not aim to capture the complete
knowledge about the application domain of football, but to provide effective means of
adapting and extending the required knowledge according to the user needs.
The suggestions of the system are based on domain-specific knowledge such as the
opponent’s team tactic and the features of the available players. Such kind of
knowledge had to be modeled through appropriate application models. However, no
such model was identified in the relevant literature. Additionally, i-footman is a
decision support system and it calls for effective modeling of human knowledge and
expertise. Hence, the system had to exploit proper knowledge technologies and
representation formalisms that would facilitate the extensibility of the knowledge
base. Consequently, modern techniques (e.g. Semantic Web technologies) were
adopted in the context of i-footman in order to represent knowledge through
expressive languages. The adoption of rules and ontologies take advantage of the
declarative programming paradigm by exploiting natural forms of knowledge.
The rest of this paper is organized as follows. Section 2 discusses the architecture
of i-footman and some modeling issues, as well. A functionality and performance
evaluation that has been performed in a simulated environment is presented in Section
3. Finally, some concluding remarks are provided in the last section of the paper.
2 System Architecture and Application Models
Two ontological models have been developed in the context of i-footman for
modeling football players and teams, accordingly. Furthermore, a large set of rules
was composed in order to express more complex concepts and relations. Two domain
experts were interviewed in order to acquire the domain knowledge captured by
models and rules1. The Pellet reasoner (v. 1.5.1) [6] that is based on Description
Logics (DLs) [1] is responsible for reasoning over the ontologies while the rule
engine Jess [4] performs the execution of rules. The selection of the reasoning
modules is based on the performance evaluation presented in [5]. The integration of
ontologies and rules is not a straightforward task, since there is no single reasoning
module that can seamlessly handle both formalisms. Hence, i-footman performs
reasoning tasks in a sequential manner and the results of the ontological reasoning are
provided as input to the rules execution process. A generic view of the framework
architecture is depicted in Fig. 1.
DL-Reasoner
i-footman
Football Players reuses Football Teams
Ontology Ontology
reuses reuses
user
Rules Rule
Engine
Formation Identification
Player Selection Tactical Instructions
Fig. 1. i-footman Architecture
1
All the application models and rules developed in the context of i-footman are provided online
in http://www.di.uoa.gr/~vpap/i-footman/
Football Players Ontology
The Football Player Ontology (FPO) has been developed in Ontology Web Language
(OWL) [2] and, in particular, the OWL-DL version. It is the first ontology, according
to our knowledge, that focuses on the description of football players. FPO provides an
extensive vocabulary referring to the various player characteristics, focusing on the
features used by a manager in order to make a tactical decision, and the tactical
instructions that a football player could follow. Some key concepts of the ontology
concern the football players (“Football_Player”), the positions they can hold during a
football match (“Position”) and the abilities of the footballers (“Ability”). Each ability
that describes a certain player is assigned a value denoting its respective quality. This
fact facilitates the definition of complex concepts through conditions of equivalence.
For instance, the following statement defines the concept of “creative” middlefielders:
fpo:CreativeMiddlefielder ≡ (fpo:hasPassing.GoodAbility
(1)
fpo:hasPassing.VeryGoodAbility) fpo:playsInPosition.Middlefielder
Football Teams Ontology
The Football Teams Ontology (FTO) provides the terms for describing the main
features of football teams and the tactical guidelines they could follow. This
vocabulary is also expressed in terms of OWL-DL and is strongly related to the FPO.
Specifically, the FTO reuses the FPO vocabulary extending it properly.
The FTO terms could be distinguished to three parts. Firstly, the ontology defines
some generic team features (e.g., the players of a team). Secondly, the ontology takes
advantage of DLs in order to classify the various team instances into certain
categories (e.g., teams that attack from the wings). Finally, the FTO models the
various tactical instructions allowing the execution of rules that will follow.
Rules
The rules constitute the declarative part of the knowledge base and are expressed in
terms of the Semantic Web Rule Language (SWRL) [3]. The basic idea behind the
adoption of SWRL was the combination of ontologies and rules in the same logical
language. The rules reuse the vocabulary provided by the FPO and FTO in order to
define more complex relationships and their structure was based on the knowledge
acquired by the experts. They can also be distinguished in four main categories2:
1. Identification Rules. These rules aim to identify the weaknesses and the
advantages of the opponent. They are mainly based on the opponent’s team formation
and the features of players in order to deduce the capabilities of the opponent. For
instance, the following SWRL rule expresses the knowledge that a team plays well
the counter attack when two or more of its offensive players are quick3:
2
For space limitation reasons, a simplified version of the rules is presented here. The complete
form of rules can be found online in http://www.di.uoa.gr/~vpap/i-footman/rules.owl.
3
QuickOffensivePlayer is an FPO concept defined through necessary and sufficient conditions.
fto:hasStartingPlayer (?t1,?p1) ∧ fto:hasStartingPlayer (?t1,?p2) ∧
fpo:QuickOffensivePlayer (?p1) ∧ fpo:QuickOffensivePlayer (?p2) → (2)
fto:dangerousAtCounterAttack (?t1,true).
2. Formation Rules. They are responsible for specifying the tactical formation that
the team will follow during a match. In particular, the number of defenders,
middlefielders and attackers is determined as well as the positions that they should
cover. The following rule describes a case where three central defenders are used:
fto:myTeamPlaysAgainst(?t1,?t2) ∧ fto:TeamWith3CentralDefenders(?t2) ∧
fto:TeamWith3CentralPlayers(?t2) ∧ fto:TeamWithSideMFs(?t2) ∧ (3)
fto:TeamWith2Attackers(?t2) → fto:playsWith3CentralDefenders(?t1, true).
3. Player Selection Rules. They make use of players’ features and tactical
formation of the teams. i-footman proposes the appropriate players to form the team’s
composition according to the classification that has been already completed through
the ontology reasoning processes. A typical player selection rule follows:
fto:myTeamPlaysAgainst(?t1,?t2) ∧ fpo:playsWith1Striker(?t1) ∧
fpo:GoodStriker (?p1) ∧ fpo:isMemberOf(?p1,?t1) → (4)
fpo:isSuggestedTo(?p1,?t1).
4. Tactical Instructions Rules. They specify the tactical instructions suggested by
the system. These rules take advantage of the opponent weaknesses and strengths
specified by the identification rules and information coming from formation rules.
The following rule denotes that if a team does not use side defenders then the
opponent should identify this weakness and attack from the wings.
fto:myTeamPlaysAgainst(?t1,?t2) ∧ fto:TeamWithNoBacks(?t2) ∧ (5)
fto:TeamWithWingers(?t1) → fto:shouldAttackFromTheWings(?t1, true).
3 Evaluation
Since no actual data and statistics were available during the development of i-
footman, the evaluation was accomplished through simulations. Specifically, the
simulation is based on two scenarios and takes advantage of two computer games that
focus on the tactical management of football teams: the Championship Manager 2008
(Eidos) and Football Manager 2008 (Sports Interactive). Such computer software
simulate a football match according to the tactical instructions that have been given as
input to both teams and generates a final result for that game. Moreover, the platforms
provided some players’ and teams’ statistics that have been inserted to the ontologies.
Regarding the first evaluation scenario, two teams with similar ratings (according
to the evaluation provided by each platform) were selected. Specifically, Barcelona
FC (Spain) was guided by i-footman while the computer handled the team of Real
Madrid FC (Spain). 40 games were simulated in each platform (80 in total). The
computer handled both teams in the first 20 games (in each platform) with no external
interference while i-footman managed the team of Barcelona in the next 20 games.
The match results of the first scenario are presented in Fig. 2. In a total of 40
matches without the intervention of i-footman, Barcelona won 14 times, Real Madrid
won 16 times and 10 games came to draw. Furthermore, Barcelona scored 45 goals
while Real Madrid scored 61 goals. Afterwards, the same number of matches was
executed with i-footman controlling Barcelona and the results seemed to be in favour
of Barcelona. Although the number of Barcelona’s wins did not increased
substantially (only by 1), the number of losses decreased by 50% (8 instead of 16).
Furthermore, Barcelona scored 51 times (instead of 45) while opponent scored 36
(instead of 61). This means that the recommendations of i-footman lead to significant
improvement of the team performance.
Barcelona FC vs. Real Madrid FC (match results)
18 17
16 CPU
16 15
14
14 i-footman
12 10
10 8
8
6
4
2
0
Wins Draws Losses
Fig. 2. Match results of the first scenario
The second scenario concerned the evaluation of i-footman when playing against
an opponent with better ratings than the one managed by the system. Specifically,
Olympiacos FC (Greece) was handled by the system and played against Real Madrid.
The results of this scenario were very similar to the former with a minor improvement
of the total number of wins accomplished by i-footman.
Generally, i-footman seems to perform well in such simulation environments.
Although the performance of the user’s team does not improve substantially, the
adoption of i-footman seems to tackle the opponent’s performance successfully. This
stems from the fact that both goals and wins of the opponent’s team are obviously
decreased. The large number of rules that refer to the defending strategy of the team
may lead to that. Specifically, the tactical formation and composition of the team are
mainly adapted to the advantages and less to the weaknesses of the opponent’s team.
This could be improved by extending and modifying the rules of the knowledge base
(possibly through interviews of more domain experts).
Regarding the time performance of the framework, the expected response time was
7741ms (as measured on a typical Desktop PC). Such time seems reasonable since it
includes all the reasoning processes (i.e. hierarchy classification, instance checking of
FPO and FTO ontologies and rules execution) that were performed.
4 Conclusions and Future Work
We have presented a knowledge-based framework able to support decision making in
the context of football. More precisely, the paper focuses on the architecture of i-
footman, its application models and the simulation results derived during the
evaluation process. The main issue that has been identified during development was
the lack of an integrated reasoning framework capable of handling ontologies and
rules. Today, there is no efficient reasoning module that can reason over both
formalisms seamlessly. Hence, the developer has to perform ontological reasoning
and provide the results as input to the rule engine and vice versa in order to achieve
effective knowledge management.
As described, i-footman was designed to facilitate the addition of possible
extensions regarding its functionality by modifying the knowledge used to describe
the application domain of football. This process would be facilitated by the
exploitation of learning techniques that target to automate the generation of the
declarative part of the knowledge base (i.e. the rules). This could be achieved either
by accessing actual statistical data about players and teams or by taking the results
that arise from the simulation platforms as real. Since no real data are available, the
exploitation of learning algorithms over virtual data seems to be more feasible.
Finally, more expressive knowledge representation languages that support fuzziness
could be adopted in order to deal with knowledge uncertainty issues.
Acknowledgements
This work is supported by the Special Research Grants Account of the University of
Athens through the Kapodistrias Programme (Research Grant Number: 70/4/7819).
The authors would also like to thank the football managers Panagiotis Lemonis
(Olympiacos FC, 2007/08) and Nikolaos Nioplias (Greece Under 21’s National Team,
2007/08) who provided basic knowledge for understanding the domain of football and
valuable comments towards the improvement of this work.
References
1. Baader, F., Calvanese, D., McGiuness, D., Nardi, D., Patel-Schneider, P.F.: The Description
Logic Handbook: Theory, Implementation, and Applications. Cambridge: Cambridge
University Press (2003)
2. Dean, M., Schreiber, G., Bechhofer, S., van Harmelen, F., Hendler, J., Horrocks, I.,
McGuinness, D.L., Patel-Schneider, P.F., Stein, L.A.: OWL Web Ontology Language
Reference. W3C Recommendation 10 February 2004. Retrieved April 22, 2008, from
http://www.w3.org/TR/owl-ref/ (2004)
3. Horrocks, I., Patel-Schneider, P., Harold, B., Tabet, S., Grosof, B., Dean, M.: SWRL: A
Semantic Web Rule Language Combining OWL and RuleML. World Wide Web
Consortium Member Submission, http://www.w3.org/Submission/SWRL/ (2004)
4. Jess, The Rule Engine For the Java Platform. Retrieved April 22, 2008, from
http://www.jessrules.com/jess/index.shtml (2008)
5. Papataxiarhis, V., Tsetsos, V., Karali, I., Stamatopoulos, P., Hadjiefthymiades, S.:
"Developing rule-based applications for the Web: Methodologies and Tools", chapter in
"Handbook of Research on Emerging Rule-Based Languages and Technologies: Open
Solutions and Approaches" (Eds. Adrian Giurca, Dragan Gasevic and Kuldar Taveter),
Information Science Reference (2009)
6. Sirin, E., Parsia, B., Grau, B., C., Kalyanpur A., & Katz, Y. Pellet: A practical OWL-DL
reasoner, Journal of Web Semantics, 5(2) (2007)