=Paper=
{{Paper
|id=Vol-2456/paper4
|storemode=property
|title=An Ontology for Ice Hockey
|pdfUrl=https://ceur-ws.org/Vol-2456/paper4.pdf
|volume=Vol-2456
|authors=Robin Keskisärkkä,Huanyu Li,Sijin Cheng,Niklas Carlsson,Patrick Lambrix
|dblpUrl=https://dblp.org/rec/conf/semweb/Keskisarkka0CCL19
}}
==An Ontology for Ice Hockey==
https://ceur-ws.org/Vol-2456/paper4.pdf
An Ontology for Ice Hockey
Robin Keskisärkkä, Huanyu Li, Sijin Cheng,
Niklas Carlsson, and Patrick Lambrix
Linköping University, Linköping, Sweden
firstname.lastname@liu.se ??
Abstract. Ice hockey is a highly popular sport that has seen signifi-
cant increase in the use of sport analytics. To aid in such analytics, most
major leagues collect and share increasing amounts of play-by-play data
and other statistics. Additionally, some websites specialize in making
such data available to the public in user-friendly forms. However, these
sites fail to capture the semantic information of the data, and cannot
be used to support more complex data requirements. In this paper, we
present the design and development of an ice hockey ontology that pro-
vides improved knowledge representation, enables intelligent search and
information acquisition, and helps when using information from multiple
databases. Our ontology is substantially larger than previous ice hockey
ontologies (that cover only a small part of the domain) and provides a
formal and explicit representation of the ice hockey domain, supports
information retrieval, data reuse, and complex performance metrics.
1 Introduction
While sports analytics in the past was limited to simple high-level statistics
based on manually extracted data, the development of new technologies (e.g.,
optical object tracking) supporting the automatic annotation of games has led to
increasing amounts of available play-by-play data, containing details about each
play event and its context (e.g., detailed game state, player positions, puck/ball
position, and timestamps). To gain a competitive advantage many teams are
already continually analyzing this data, looking for an edge on their competitors.
Today, play-by-play data and other statistics are provided by many of the
major ice hockey leagues, including the National Hockey League (NHL) in North
America (US+Canada) and the Swedish Hockey League (SHL). There are also
public websites that present statistics based on such data in human friendly for-
mats; e.g., Corsica (http://corsica.hockey) and Natural Stat Trick (http:
//www.naturalstattrick.com). These sites typically show a limited range of
performance metrics, and cannot support complex query requirements or de-
tailed insights of play-by-play data.
Ontologies provide a formal and explicit representation of the domain knowl-
edge, which can greatly benefit information retrieval and data reuse, and support
??
Copyright c 2019 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
advanced performance metrics, such as those proposed in recent research on ice
hockey analytics (e.g., [3,4,1,2,5]). Prior work focusing on ice hockey ontologies
is very limited, and existing ontologies cover only a small part of the domain.
In this paper, we present the development of an ice hockey ontology that
extends the coverage of previous efforts, map play-by-play data to the Resource
Description Framework (RDF), and validate that our solution easily can be
used to effectively answer example questions (otherwise not easily accessible)
using SPARQL queries. The ontology enables semantics-based access to existing
data, as well as the integration of different data sources. In addition to being
available on the web, the ontology will be used in applications related to ice
hockey analytics and data visualization in cooperation with a professional ice
hockey team. The design and development of the ontology and its current state
are discussed in Sections 3 and 4, respectively.
2 Related work
There is little ontology-related work focusing on ice hockey. The International
Press Telecommunications Council, which develops industry standards for the
exchange of news data, has developed a sports ontology (https://iptc.org/
std/SportsML/3.0/documentation/). The ontology defines general sports-related
concepts. While the ontology benefits from many concepts being shared across
sports, some terms end up being overloaded, leading to the exact interpretation
often being dependent on the actual sport under consideration. The ice hockey-
specific part of the ontology deals with traditional player and team statistics,
and simple event states related to power play and scoring. Similarly, BBC devel-
oped a lightweight ontology (https://www.bbc.co.uk/ontologies/sport) for
representing sports events with a focus on the organization of competitions. Fi-
nally, DBpedia (https://wiki.dbpedia.org/) contains some ice hockey-related
terms such as ice hockey league and ice hockey player.
3 Ontology design and development
The design and development of the ontology included four high-level steps: (1)
description of use cases, (2) specification of competency questions, (3) formal-
ization, and (4) validation. The process was implemented in an iterative fashion,
with refactoring, revisions, and refinements to ensure that the ontology was ex-
pressive enough to capture the competency questions.
Use cases: The use cases aim at providing ice hockey knowledge to general
users and to support professionals in the domain, including head coaches, play-
ers, general managers, and team scouts. This includes role-specific uses cases
requiring support for advanced data analytics of play-by-play data.
Competency questions: According to the use cases, a set of competency
questions (CQs) were specified and categorized as either game related, event
related, or performance-metrics related. Examples of representative CQs were:
(1) How many games end during the regular-time period?, (2) When did the
Fig. 1. Overview of the ice hockey ontology.
winning goal happen for a specific game?, and (3) What is the faceoff winning
percentage in the last X games of a specific player?. The CQs were used both
to provide the scope of the ontology, and to provide some way of validating the
ontology with respect to the use cases.
Formalization: Based on the use cases and CQs, we conceptualized ice
hockey related concepts, starting from NHL play-by-play data and the NHL
rule book. We then used OWL to formalize the ontology in Protégé. Starting
from a set of general concepts in ice hockey (e.g., game, event, person, team), we
extended the concept hierarchy by specializing these concepts. For example, the
event concept was specialized into penalty event, action event, etc. Furthermore,
we defined class properties and constructed semantic relationships.
Validation: We validated the ontology using the OOPS! service, the HermiT
reasoner, and RepOSE. We then mapped the play-by-play data to RDF using
the RDF Mapping Language (RML), converted the data to RDF, and provided
validation tests for each CQ using one or more SPARQL queries.
4 Current coverage
The current version of the ontology can be used to represent: (1) basic knowledge
about the ice hockey domain, (2) game events and game sequences, and (3)
describe the game context of events.
The ontology currently contains 125 concepts, 100 relations, and 892 axioms.
Figure 1 shows an overview of the concept hierarchy, and a detailed description
of Shot-event. The general concepts covered in the domain include, for example,
Arena, Game, League, Penalty, Period, Person, Team, as well as concepts on
the event level such as Game-event, with Action-event and Faceoff-event as sub-
concepts, and Game-state to represent the event context.
Shot-event is defined as a sub-concept of Action-event which in turn is a
sub-concept of Game-event. As shown by the axioms on the right-hand side of
Figure 1, we define a Shot-event from the perspective of the attacking team,
and include the shooting player and the shot type (e.g., slap shot). The bottom
half of the right-hand side of Figure 1 shows the axioms that Shot-event inherits
from Game-event. The first axiom specifies that a Shot-event has a specific game
context (Game-state), capturing the set of players on the ice for both the home
and away team, and the team disposition, for example.
Mapping play-by-play data to RDF: The amount and quality of match-
specific data available for ice hockey varies greatly between different leagues,
and the features reported depend largely on the type of systems employed. Typ-
ical datasets referred to as play-by-play data include a discretized representation
of ice hockey games, where timestamped events deemed relevant for post-game
analysis have been recorded along with some contextual information. Also the
formats used to represent such datasets differ greatly, varying from fully nor-
malized database dumps with foreign key relations, to tables of semi-structured
data. In this paper, we focus on the representation of (both public and private)
datasets provided for NHL and SHL, and provide RML mappings to declara-
tively capture the data based on the proposed ontology.
Evaluation: After constructing the ontology and creating mappings from
play-by-play data to RDF, we validated the ontology by answering CQs related
to the use case. For each CQ, we created one or more SPARQL queries to validate
if the ontology captured sufficient information to answer all CQs.
5 Conclusion
We have presented ongoing work on developing an ice hockey ontology that con-
ceptualizes general ice hockey domain knowledge and events in play-by-play data.
Based on the proposed ontology, we provided a mapping of play-by-play data to
RDF using RML, and validated the ontology against a set of SPARQL queries
solving competency questions derived from different role-specific use case. The
ontology provides a formal and explicit representation of the domain knowledge
that supports information retrieval, data reuse, and can help in the retrieval of
more advanced performance metrics from play-by-play data in ice hockey.
References
1. Liu, G., Schulte, O.: Deep reinforcement learning in ice hockey for context-aware
player evaluation. In: IJCAI. pp. 3442–3448 (2018)
2. Ljung, D., Carlsson, N., Lambrix, P.: Player pairs valuation in ice hockey. In: Ma-
chine Learning and Data Mining for Sports Analytics, LNCS 11330. pp. 82–92 (2019)
3. Macdonald, B.: An Expected Goals Model for Evaluating NHL Teams and Players.
In: MIT Sloan Sports Analytics Conference (2012)
4. Routley, K., Schulte, O.: A Markov Game Model for Valuing Player Actions in Ice
Hockey. In: UAI. pp. 782–791 (2015)
5. Sans Fuentes, C., Carlsson, N., Lambrix, P.: Player impact measures for scoring in
ice hockey. In: MathSport International. pp. 307–317 (2019)