Searching for similar football possessions is typically carried out by reviewing match videos by a football analyst, which is time-consuming and prone to inconsistencies caused by human subjectivity.
In this master's thesis, we designed, implemented, and presented a solution for automatically searching for similar football possessions using information retrieval methods. After algorithmic preprocessing, sequences of events are combined into possessions and converted into text documents, which are then indexed. A domain expert determines the document and index schema, as well as possible attribute weighting. The system supports filtering, various types of queries, and the display of contextual metadata. The developed solution thus enables fast and accurate searches for tactically similar possessions without the need for time-consuming manual labelling or complex supervised learning.
Evaluation on data from the Slovenian First League demonstrated high efficiency and short response times, and the approach is scalable. The developed solution enables football analysts to recognise game patterns more quickly and objectively, representing a contribution to data-driven sports analytics.
|
