Skip to content. | Skip to navigation

Informatik 5
Information Systems
Prof. Dr. M. Jarke
Personal tools
You are here: Home Theses Event Based Activity Recognition In A Virtual Multiplayer Team Game


Prof. Dr. M. Jarke
RWTH Aachen
Informatik 5
Ahornstr. 55
D-52056 Aachen
Tel +49/241/8021501
Fax +49/241/8022321

How to find us

Annual Reports





Event Based Activity Recognition In A Virtual Multiplayer Team Game

Thesis type
  • Master
Student Vladislav Supalov
Status Finished
Submitted in 2013
Proposal on 15. May 2013 16:45
Proposal room Seminarraum I5
Add proposal to calendar vCal
Presentation on 27. Sep 2013 00:00
Presentation room Seminarraum I5
Add presentation to calendar vCal

Competitive videogames, also known as e-sports, is a young and growing field, gaining increasing popularity across the demographic worldwide. Unlike traditional sports like soccer or baseball, most of the games do not have established metrics and automatic analysis methods, and affords many ways of incuiry. The focus of this thesis is the application of Activity Recognition and Player Modeling techniques to match-replays of an online multiplayer videogame: DotA2 - a team game with great depth, complexity, an active competitive scene and a large player base. In the first step, a web-based match annotation platform is developed and provided to the community. Through the usage of the platform, a dataset of annotated matches is assembled in a typical crowdsourcing fashion. Existing Activity Recognition methods, utilized in Pervasive Computing for the interpretation of sensor data are adapted to event based data, trained and evaluated on the dataset, with the goal to distinguish between different characteristical players behaviours in an automated and reliable fashion. In the final step, the developed system is integrated into the previously developed annotation platform, to provide semantic understanding of new matches on demand.

The focus of this thesis, is to apply Data Mining and Machine Learning techniques to match replays of an online multiplayer videogame: DotA2. The main task is to develop a machine learning approach, which can classify different player behaviour based on previously observed game events, thus providing automated game understanding. The acquisition of a relevant dataset and the subsequent utilization of the created approach in a productive fashion are necessary subgoals of this thesis. The main steps toward a solution are the following: 1. A match-annotation platform where users can highlight and comment on specific situation and player actions. The thus produced annotations will be used as dataset for the subsequently applied machine learning approach. 2. A robust processing pipeline has to be established. The replay format the low level encoding of game events has to be processed and promising features have to be identified and enabled to be extracted in an automated fashion. 3. A method for player activity recognition needs to be derived from common sensor-based activity recognition approaches to be applied to the event-based data and evaluated on the dataset. 4. Integration of the developed method into the original match-annotation platform, to enable a productive utilization of the created system.

Related projects

Document Actions