Abstract

The collection and analysis of videogame players' actions in the game world, known as
game telemetry, is a common technique for understanding the behaviour of players. This
process, known as Game Analytics, often uses data visualisation to allow designers to
manually analyse features of the data. Heatmap visualisation, a grid-based visualisation
showing how often an event occurs across the game world (e.g. �ring of weapons), is
used widely in the games industry for visualising aggregated data, but has limitations
when used to classify player behaviour at an individual or group level. Existing works
using clustering to identify player behaviour yield results that must be interpreted by an
expert, a problem acknowledged by existing research. Motivated by these limitations,
this work presents the novel application of dendrogram visualisation as a means to
interpret large datasets of heatmaps, through the use of hierarchical clustering, to aid
designers in exploring and analysing player behaviour. This allows an intuitive and well-
understood visualisation technique (heatmaps) to be used for cluster analysis, presenting
intelligible results to a game designer, in a format they are familiar with.
To evaluate dendrograms as a design tool, a system was designed and implemented to
visualise player data, using heatmaps, with hierarchical clustering being performed on
these heatmaps, the results displayed as a dendrogram. A feasibility study was con-
ducted with a set of game designers, to understand the opportunities and limitations
of dendrograms as a game analytics tool. The results a�rmed the utility of heatmaps
for visualising aggregate data, but visual complexity increases in large quantities. Den-
drograms were found to be initially di�cult to read, but showed promise for analysing
large sets of data and guiding the designer to interesting areas of the data, provided
they could \drill down" into the base data (heatmaps). In light of these �ndings, a us-
ability study was designed and conducted with a set of 40 game development students,
where they were presented with realistic game design scenarios, and asked to �nd an-
swers to analytics questions using heatmaps and dendrograms. The results showed that
whilst dendrograms were initially di�cult to understand, they were used to successfully
explore and understand cluster relationships, with participants providing the correct
answers grounded in the data. Furthermore participants reiterated the need to explore
the base data (heatmaps) to understand the cluster relationships of the dendrogram.
This work concludes that dendrograms represent a viable and useful tool for identifying
interesting behaviour patterns within a heatmap dataset. Whilst some familiarity is
required with the tool, it is possible to use dendrograms to explore behaviour clusters
within a large dataset, and this work presents a solution to the limitations of analysing
player behaviour through the use of heatmaps in large datasets. This work highlights
a number of avenues for future work, such as deploying and studying dendrograms in a
game production setting, or evaluating the dendrogram visualisation in di�erent game
genres.