Alternative title: Rapid Skill Capture in a First-Person Shooter
Abstract: Although some datasets exist that concern player input to a video game, these either use unconventional modes of input or concern trivial games. This dataset provides full keyboard and mouse data for a non-trivial, commercial-quality first-person shooter. Designed to explore skill capture in a video game, it also includes several high-level data such as kills, shots fired and points scored. In addition, it contains some basic affective responses to each of the games. Participants in the experiment alternated between playing a 3 minute deathmatch on a desktop computer and answering questions about their experience of the game. Every second game participants filled in an additional questionnaire to compare their previous two games. This dataset contains 476 games from 45 different players over two experiments that ran between 2013-02-17 to 2013-03-01 and 2013-12-02 to 2013-01-22. The data provided also includes information about the experiment to aid reproducibility, and some helper scripts for managing the data.
Table of contents: The data is provided in both JSON and Python object format for convenience. The former is human readable, and is more flexible. The Python objects, on the other hand, can be loaded directly into Python and used with the supplied scripts. Each game is stored as a separate file and are bundled together into .zip and .tar.gz archives. These contain a list of events for each game along with any custom game settings, such as the difficulty. The users file contains a list of all users (identified by an ID) along with their answers to the demographic questionnaire, such as age, or the number of hours played. The sessions file contains the responses to the post-game questionnaires and links to the games using an ID. For the Python objects described above, the autoclass package is provided, which primarily contains the classes for each type of object. In addition, a datamanager class is provided that loads the object from file. An example.py file has been included to demonstrate how best to use the scripts. Each player was presented with an information sheet and tutorial, which have both been provided as PDFs. The configuration file used for the experiment has also been provided.
Other: There were three games for which the first few seconds of data are missing. This occurred when the participant started the game before they finished filling in the questionnaire. The experimenter would then direct them to fill the questionnaire in before returning to the game. The users and the missing data are (with the Game ID in brackets):  User 13, 3rd game (Bath): 18.612s,  User 44, 4th game (Bath): 16.811s,  User 30, 8th game (deli): 21.166s. The bots that the user faced were, in the analysis, assigned TrueSkill estimates. These are the estimates that were calculated for each of the bot ranges, with sigma in brackets: 40 - 50: 18.944 (0.639), 50 - 60: 20.776 (0.638), 60 - 70: 20.593 (0.638), 70 - 80: 23.685 (0.637), 80 - 90: 26.342 (0.638), 90 - 100: 31.441 (0.638).
Date of data production:2014-01-22
|Date made available||4 Apr 2015|
|Publisher||University of Manchester|