Sponsored By

Player profiling in practice

While player modelling has seen its use in games, player profiling is relatively new and many of its uses are still being explored. This article aims to give the reader an overview of the ways player profiles can be used and created.

Kevin Hutchinson, Blogger

April 14, 2016

9 Min Read

This article aims to give the reader an overview of the ways player profiles can be used and created.

While player modelling has been the subject of research for quite some time and has seen its use in games, player profiling is relatively new and many of its uses are still being explored.

What is player profiling

Player modelling is the process of gathering player data to create a model of their behaviour. Player profiling is a subcategory of player modelling that focuses on information that is not game specific and remains static over longer periods of time. Examples of data typically included in a player profile would be personality, cultural background, gender, and age.

Because player profiles are not game specific, models created in one game can be used in another. Additionally, models could be created by combining player data collected from different games or outside of games. This ‘portability’ means player profiles are more suited for predictive use rather than adaptive use like player models.

To give an example, player profiles imported from other games could allow a game to be pre-adjusted to give a new player the best chance of a positive experience from the first moment. Player modelling would need time to establish a model of the player’s behaviour and adjust the game accordingly.

This ‘portability’ used to be of minor significance as games were usually isolated from each other; however, with the ever increasing popularity of online gaming platforms such as Steam, Facebook, and app stores, player profiling is going to be a larger factor of audience analysis in the future.

Before we continue and look at the uses of player profiles we should not ignore its social implications. It must be stressed that any data gathered needs to be used responsibly and with respect for the players. While the ethics of player profiling is definitely a topic worthy of discussion, this article will focus on the technical aspects of player profiling.

Uses of player profiling

Player profiles share many of the uses of player models, but with some unique advantages and disadvantages. Player profiles can be used for:

  • Dynamic difficulty adjustment. Player profiles can be used to adjust the difficulty of a game to an individual player’s abilities. By adjusting numbers and placing/removing obstacles, players are challenged enough to keep them interested while avoiding frustration. As mentioned before, this profile could be imported from other games to give the most enjoyable difficulty at the first play.

  • Content generation. Player profiles can be used to procedurally generate personalised content or guide the manual creation thereof. By changing the types of challenges a game or level provides, based on a player’s profile, we increase the chance of it eliciting the desired player experience.

  • Improving AI. Allow the AI to mimic or adopt player behaviour to improve itself or become more human-like.

  • Monetisation. By presenting a player with content they are likely to buy, player profiles can be used for monetisation purposes. This is especially useful in free to play games. Player profiles can be used to make personalised offers, for targeted advertising, to change the way content is presented in a marketplace, or even to guide the developer in choosing new content to add to the game based on what is likely to sell.

    Player modelling does this by looking at the player’s previous acquisitions and what content they use the most. Player profiling on the other hand might look at what type of player someone is and what kind of content this type of player is likely to buy.

In addition player profiles have a unique use:

  • Attribute testing. Because player profile data is not game specific the information can be useful outside the game. An example of this would be personality profiling. This is traditionally done with a Likert-scale test.

    However, this explicit way of testing means subjects know they are being tested and the results will be influenced by unconscious processes like self-deception, image management, or simply the lack of self-knowledge.

    In the future, games might be able to measure implicit attributes by avoiding the subject’s filters. (For further information on implicit tests, see [1] and [2])

Using and gathering data

Using data

The most common way of using player profile data is categorizing players in groups. Groups are often predetermined and associated with certain assumptions. These assumptions are then used to impact the game.

Let’s say we are developing an educational game that helps children learn to read. If we have a player profile that includes their age we could group the players by age. We could then assume children in a certain age group have roughly the same reading skill and balance the game based on this data.

It should be noted that the chosen grouping method has a large impact on how the data can be used. In the example above we might group the players in the following age groups:

  • Ages 0-3

  • 4-5

  • 6-7

  • 8-9

  • 10+

However, if we were developing a shooter and wanted to change the amount of gore based on player age this grouping would be useless and we would be better off distinguishing groups somewhat like these:

  • Ages 0-14

  • 15-16

  • 17-21

  • 22-30

  • 31+

This way of generalization makes our data easy to use but comes at a cost. Even if the groups are set up perfectly, people are unpredictable; there are always outliers. When you base decisions on wrong assumptions… Things go wrong.

When categorizing players you should always ask yourself: “What are the consequences of a wrong assumption?” and if the consequences are minor “Is it acceptable for this assumption to be false for x% of players?”. If the answer is ‘no’ to either question then you can narrow down the groups (possibly by using more attributes) or try a different approach.

Going back to the children reading game example: we grouped children with an age difference of 2 years in a single group and assumed their skill was similar. However, children learn quickly and what a 4-year-old finds challenging will be easy for a 5-year-old.

We chose to narrow down our groups to solve this problem. Now we only assume children of the same age are equally skilled at reading. Again our assumption is wrong: children learn at different speeds. In this case grouping by age might not be the right approach at all and, albeit harder to do, we might be better off trying to model a player’s skill at reading directly.

Players can be grouped by personality by using personality types. By grouping a range of personalities under a personality type we can then attach assumptions to these types. The most common grouping in games would be Bartle’s personality types. Bartle distinguishes Killers, Achievers, Socializers and Explorers by rating players on 2 independent scales.

385px-Character_theory_chart.svg.png

Bartle's scales and player types 

While Bartle’s personality types (and its variants) are meant specifically for games, there are alternatives that are not. Most notable amongst these alternatives are the Myers-Briggs Type Indicator and Keirsey Temperaments.

No model of personality types has been universally accepted. New ones and variations/combinations of old ones are still added. In practice the model that is used is often determined by its purpose.

Gathering data

Having discussed the uses and setup of our data, one question remains: how do we get the data? There are essentially 2 ways to gather data for player profiles:

Ask. Simply ask the player for the data you need. This will often work for basic information like age and gender but any sort of survey will scare off players. App stores avoid this by allowing apps to request player information: if the information was previously provided the player only needs to accept. In any case, the types of data that can be gathered this way are limited.

Alternatively, we can model it from player behaviour. This will work for any kind of data as long as a method can be found to model it. This is where the problem lies: we don’t know how to model most types of data. How do we determine someone’s age by analyzing their gameplay? Or their personality? Establishing relations between gameplay and player attributes is a growing topic in research, albeit one that is also sometimes controversial.

In 2011 Lankveld, Spronck, Herik and Arntz [3] successfully modelled the 5 traits in the ‘Big Five’ personality model by analyzing conversation choices in a role-playing game. This proves there is indeed a relationship between player behaviour in games and their real-life personality. While the very specific game environment means the method of modelling personality can’t be used in other games, it opened the door for research to find more general methods that can.

Conclusion

Player profiling has a myriad of uses, many of which are not, or cannot be, fully exploited yet. Player profiling can be a useful tool to improve the player experience of a game. However, we should be aware of its potential use for malicious purposes and work to avoid this.

Research is still in the process of proving there are connections between gameplay and player attributes. When this is followed by exploration of the best way to model these attributes, we can expect player profiling to see widespread use in games.

_______________________________________

About the author

Kevin Hutchinson is a Masters student at the NHTV in Breda, the Netherlands. He is a programmer specialising in procedural generation. Currently writing his thesis, he is researching the modelling of self-esteem from player behaviour in games.

References

  1. W. Hofmann, B. Gawronski, T. Gschwendner, H. Le, M. Schmitt, A Meta-Analysis on the Correlation Between the Implicit Association Test and Explicit Self-Report Measures, 2005

  2. A. Greenwald, S. Farnham, Using the Implicit Association Test to Measure Self-Esteem and Self-Concept, 2000

  3. G. Van Lankveld, P. Spronck, J. Van Den Herik,  A. Arntz, Games as personality profiling tools, 2011

Read more about:

Featured Blogs
Daily news, dev blogs, and stories from Game Developer straight to your inbox

You May Also Like