Tweeting GDC
by Jesper Juul, Mia Consalvo,
John Sharp & Ulrika Bennerstedt
What do people talk about on the
backchannels of the 2009 GDC conference?
What do people talk about on the backchannels of the 2009 GDC conference?
Based on a shared interest in the public discourse around conferences, our Real Time Research group decided to investigate the “backchannel” of the 2009 Game Developers Conference (GDC). We chose to focus on the micro-blogging service Twitter and the stream of commentary generated using the #GDC and #GDC09 hashtags. Because we anticipated a large amount of data and had little time to review and analyze, we recognized that the traditional data analysis tools we planned to use, including manual analysis and bar chart visualizations would not be sufficient given the large quantity of data captured and the short time three-day time frame for the research project. We therefore targeted Wordle as a powerful means of presenting the data drawn from the stream of tweets captured over the five-day period of our research project.
Methods
As the data that we gathered are based on short (up to 140 characters) text-based postings on the micro-blogging service Twitter (http://twitter.com/), we needed to manage these to answer our research question. In order to do this we started to define which posted messages, also known as tweets, should be included in our research. A common practice by conference participants is to include in tweets related to a conference in order to make clear that their tweet was about the given conference. In the case of GDC 2009, there were two commonly used hashtags: #GDC and #GDC09. Those interested in following the Twitter backchannel of the event knew to search Twitter using these hashtags in order to follow the tweetstream for the conference.
We focused on tweets including the #GDC and #GDC09 hashtags in our data mining. In this way we used Twitter’s own search capabilities to pull out tweets that were tied to #GDC and #GDC09 during the week of the conference (Monday-Friday). We sorted the tweets by day to better account for initiation and development of topics, and to see if events that took place at the conference were picked up by other conference participants. The amount of tweets made each day shifted, but an average of 825 of tweets was found per 24-hour period.
With that data, we performed two analyses. Firstly, an analysis of tweets specifically referencing the GDC keynote by Satoru Iwata was performed, with results presented using traditional methods (bar charts/graphs). Secondly, a parallel analysis was made by means of Wordle, a visual tool (http://www.wordle.net/) that counts frequencies of words and then represents them in aesthetically appealing “word clouds”. In this process the tool makes more recurrent words appear larger in the cloud, allowing size to correspond with frequency. The end result can be manipulated by changing colors and the form of the cloud. By means of these word clouds, we then made some preliminary analyses based on what we found to be relevant ways of explaining the data.
Findings: Analysis of
Iwata Keynote Tweets
Instead of hunting down conference attendees and making them answer a survey or asking participants directly about what topics they initiated and discussed most frequently at GDC, we studied the participants’ conference related practices on the social networking service Twitter.
The first analysis was concerned with exploring how participants twittered about the first keynote of the conference, given by Nintendo President Satoru Iwata. First, we wanted to find if attendees twittered at all about Iwata’s talk, and also how they did so. We looked for tweets that included either the tag #GDC or #GDC09, and also made some mention of “Iwata,” “Nintendo” or the first keynote talk. We explored how they twittered about the content of this talk (including journalistic accounts of what he said, positive and negative reactions to it, jokes made about it, and the like); and we also investigated how participants employed various functions of Twitter such as re-tweeting, posting links, and responding to others’ tweets. The results are presented below.
As seen in Figure 1, the greatest amount of tweets concerned anticipation before the actual keynote began. Many tweets mentioned waiting in the (long) line that was forming well before the event began. Some mentioned curiosity about what would be discussed, while others merely noted the wait time they were enduring. The next greatest amount of tweets focused on summarizing or vpresenting information gleaned from Iwata’s keynote. These tweets were not evaluative, but instead merely were repeating information that Iwata was giving out. To a far smaller degree, tweets offered positive or negative reactions to the keynote address. It seems that more twitterers were interested in passing along information, rather than giving responses or evaluations of the keynote.
Twitterers also highlighted a few notable areas from Iwata’s talk (see Figure 2)—particularly his mention of a “death spiral” that could occur in the game development business, a giveaway of the DS game “Rhythm Heaven” to promote its impending release, the next Zelda game, and discussion of game designer Shigeru Miyamoto’s particular design process. The death spiral in particular was well received, as developers quickly posted pictures of Iwata’s slide of the spiral, with one creative individual Photoshopping a game CD box to promote the Nintendo game ‘Death Spiral’.
Finally, as seen in Figure 3, we looked at how twitterers employed the particular mechanics of Twitter to direct their tweets in particular ways. Somewhat surprisingly to us, most tweets were basic, with only a small number employing retweeting or replies to other tweets. Several more tweets included links to websites, but again, most tweets referred to themselves, and did not appear to be part of a larger, specific conversation.
Findings:
Wordle Representations
In our second analysis, we were inspired by the way the tool Wordle can be used to visually represent word frequencies. We developed a small Java program that queried Twitter.com for tweets about the Game Developer Conference. We then removed sender names and times from the tweets before processing them in Wordle. By analyzing discrete sets of tweets from individual conference days, we found the word clouds generated by Wordle to be a fruitful presentation tool. Wordle uses basic word counting techniques to generate word cloud visualizations that use size to indicate the relative frequency of word usage within the given data set. Below we present the five different days based on the restrictions on tweets signed with the tag ‘#GDC’.
The first day of GDC, which is comprised of the niche summits (i.e. Mobile, IGDA Education SIG, Indie Games, etc.) saw a general focus of tweets on topics related to the conference theme of games but also including a more diffuse range of topics such as iPhones as well as on the fact only summit sessions were taking place.
On the second day of the summits, the words used mostly by the twitterers were again the words games and game, but another word emerged as just as common; party. One explanation about the word party is that it was used in the context of social events outside the conference. The conference attendees are either recurrently twittering about a certain party or about different parties; based on this data set, there is no way to shed light on which. Words as tomorrow (seen above the word game) indicate some prospective topics about what’s to come, highlighting future events of importance.
The tweets about the summits began to be outnumbered by the tweets about the conference proper, which began on Wednesday. This stands to reason, as the summits are much more lightly attended.
Figure 6. The third day (Wednesday).
Substantially different content of tweets on the third day of the conference is seen in the graphical model of the third day’s word cloud. Here words such as Nintendo, Iwata and keynote are the most common. This can be understood as a situation in which participants took up for discussion what was happening from the conference website and the subject matter of that day’s keynote. In relation to the day before, the word party has receded in importance (seen above the word iwata).
Figure 7. The fourth day (Thursday).
In Figure 7, another keynote presenter is brought up on Twitter by the word Kojima (Hideo Kojima, designer of the Metal Gear series). Here the word party is seen in a smaller text beside the word Kojima, indicating that participants were planning social events or commenting on earlier ones. Here the twitterers make the ‘last day’ a topic, besides game designers such as Wright (Will Wright).
Conclusions & Next Steps
The use of two data analysis and visualization techniques proved to be instructive. The use of traditional analysis methods for the tweets relating to Iwata’s Wednesday keynote—where we as researchers structured the data, labeled emergent themes, and made inferences about what various frequencies might suggest— and the more open-ended use of Wordle to visualize the complete data set suggested different uses. The presentation of the Iwata data as interpreted by our project team lead to a focus on the part of our audience on our methodologies rather than the data.
On the other hand, the word clouds, which presented the full data set, led to the audience joining us in the analysis of the data.. Using Wordle to provide an initial structure to the Twitter data was inspiring for us as researchers. As a visual data mining tool, word clouds provided an excellent method for drawing to the surface trends that otherwise might be overlooked. These visually translated statistics (i.e. word frequencies) seems to inspire people to become more engaged in active interpretation themselves than compared to traditional graphs in which the researcher controls the interpretation.
In order to tune in to what people at the conference were talking about, we went online and studied conference attendees’ text-based postings on the social networking service Twitter. As this tweeting practice is only done among certain members of the conference attendees, our analysis is limited to a certain population. It can be interpreted that what goes on there might be read by more than the messages being posted. The postings in themselves can also be seen as a way to talk about how people make themselves and their ideas/thoughts/experiences/opinions ‘heard’ and ‘seen’ online by a specific community (i.e. GDC).
More generally, this work makes visible methods for utilizing social interaction in already established social media and ways to work with such forms of computer-mediated communication using both traditional research methods and visually inspiring word-counting tools. We see the social network services such as Twitter not only as a new and exciting way of gathering data but also as a way to follow what participators on conferences actually say and do. For some, such communication is everyday practice; for others, it might be viewed as an exotic data-mining excursion.
As for the way we chose to present and analyze the participants doing there are some positive as well as negative aspects. Wordle functions as both a data-mining tool, by letting us process chosen amounts of words, and as a visually-appealing presentation tool. As has been brought up earlier, the traditional use of graphs when presenting our findings for an audience give rise to certain expectations in the audience. This can be caused by the strong tradition within the research community of using particular forms of graphs, making such standard representations a core part of the researchers’ toolbox, often undisputable. The word clouds might be interpreted as representations that are more open for interpretation, where the researchers do not have the final say on the interpretation. The word clouds of the conference attendees most recurrent topics opened up for meaning-making practices when presenting out result for the GDC audience that were not restricted to the stories that we, the researchers, in the group presented. The clouds aesthetically appealing appearances seems to engage people in a way that they overlook the fact that it is, at root, statistics. That is, the word clouds seem to engage people who might not usually attempt to unpacking statistical data on their own.
Using Twitter together with Wordle becomes a first step into statistics. The word cloud acts as an illustration of something, a not yet analyzed phenomena. Presenting our word clouds for the audience at the conference gave rise to other explanations of the data, other stories being told. This has the consequence that these meta-level interpretations of the words could miss out on the function of the word in its original context. In other words, the same word could be posted in very different situations, making the word have different meanings. This can of course be overcome by going back and studying the details of how the word is used, for example how the word party is used on a specific day in their various postings. However, the word clouds open up for many storytelling events that makes it a tool for quickly getting a survey of the topics in circulation among a group or community of people.
One consequence of using Wordle, then, is that the interpretations, or stories, that the word clouds represent are dragged out from the context they were made in. Thus, interpreting them will always be an imagined way of putting them back in some imagined context, particularly in relation to the other recurrent words in the clouds. From our perspective, we can present a story about what we see in the clouds that, if we don’t consider the context the word was originally used in, might be strange from the Tweeters’ point of view.
Postscript
Our research project has lead to a spin-off project by Local No. 12, a game design collective made up of Mike Edwards, Colleen Macklin (RTR alumni), John Sharp (member of the original project team) and Eric Zimmerman (one of the organizers of the RTR project). Working with the idea of mining conference related-tweets, this group designed and developed the conference game Backchatter. Sharp and Zimmerman saw the potential to develop a game around the tradition of conference reporting through Twitter in order to more fully realize the value of the backchannel reporting.
To conclude the game, the game’s creators hold a conference session in which they use Wordle to present the data set. With Backchatter, two sets of data are presented in the word clouds: the words Backchatter players anticipated would be tweeted, and the actual words tweeted during the conference. As happened at the RTR presentation, audience members join in the interpretation of the visualized data set.
Backchatter was playtested at the 2009 Games for Change Festival in New York, and then premiered at the 2009 Games, Learning and Society conference in Madison, Wisconsin and ran at the Digital Games Research Association conference in September 2009 and the Indiecade Festival conference in October 2009.
Play Style Survey
by Tone Vold, Richard Marzo & Annika Waern
Is there any coherence in how different
professions place themselves as players on the Bartle’s graph of different play styles?
Is there any coherence in how different professions place themselves as players on the Bartle’s graph of play styles?
At the beginning of the RTR workshop, we were given some choices for theory, topic, and method to work with and, after some swapping and discussion during both the first session and later meetings, we decided to use the topic card on Play Styles which depicted Bartle’s (1996) Interest Graph. Our goal was to see where the participants at GDC09 would place themselves as gamers and whether there were any differences among participants based on occupation. Would, for example, programmers always place themselves as “interveners” or “achievers”? Would managers be “Socializers”?
Theoretical Framework
The Interest Graph was developed and presented by Richard Allan Bartle, a British writer, professor and game researcher. He has also co-authored the first Multi-User Dungeon (or MUD) (“Richard Bartle,” 2009). Bartle found that there were four things that gamers enjoyed about MUDs: (1) achievement within the game context, meaning that they gave themselves goals within and related to the explicit goals of the game; (2) exploration of the game, meaning that they wanted to explore the virtual world that this MUD provided; (3) socializing with others, meaning that they used the game to get in touch with and communicate with other players; and (4) imposition upon others, meaning that they wanted to compete or otherwise interact with others either in combat or otherwise.
Thus, based on Bartle’s (1996) framework, one can categorize gamers as achievers, explorers, socializers, or interveners. Whereas the achievers are interested in acting on the world and mastering the game, the explorers want to be surprised by the game and interact with the world, the socializers want to interact with other players, and the killers/interveners want to act on other players. This results in the graph where the X-axis goes from interest in players towards the right to the environment. The Y-axis represents the differences in “acting with” at the bottom to “acting on” on the top (Figure 1).
Method
We decided on making a board (Figure 2) and have conference participants place post-it notes as to where they see themselves as gamers. Participants were asked to choose a color of post-it note that would best represent their occupation using the following categories developed for this study (Table 1).
We started out with the blank board with only the interest graph drawn on to it and walked around in the convention area and stopped participants and asked them to pick a post-it note that would best represent their occupation, write their job title on it, and place it on the board in the quadrant corresponding to how they would characterize their game-play style. We carried the huge cardboard around from table to table, asking participants to take part in our little survey. Surprisingly, very few turned us down and most people were very positive and took time to respond properly (see Figure 2).
Results
The results from our research were quite interesting. We had a total of 66 respondents. The distribution of profession category is shown in Table 1.
Across all responses, there were few self-reported ”interveners.” Programmers placed themselves “all over the place” with most tending towards the ”focused on world” end of the horizontal axis (in contrast to “focused on people” end). Audio and visual professionals classified themselves as ‘explorers’, that is, placed themselves more toward the ”focused on world” end of the horizontal axis and with more emphasis on ”interacting with” (bottom of vertical axis) rather than” acting on” (top of vertical axis). Business and management professionals gravitated towards both the “socializers” quadrant (interacting with players) and the “achieving” quadrant (acting on world). Participant who chose the category “other” were relatively evenly distributed between the “socializers” quadrant and the “explorers” quadrant (interacting with world) with only one exception.
Discussion
Despite our relatively small sample size (only 66 out of all GDC-participants), we did see some trends regarding profession and play style. The general trend towards an interest in ‘worlds’ rather than ‘people’ is perhaps the most interesting observation. It makes sense that developers and artists would have a high interest in worlds, since so much of the effort in creating a game must go into the world simulation; ranging from physics engines to visuals. The trend was also particularly pronounced for audio and visual artists, who tended to classify themselves as ‘explorers’. On the other hand, business and management professionals had a tendency towards classifying themselves as socializers and achievers, again roles that rhyme well with their chosen profession.
The fact that so few participants chose to classify themselves as ‘interveners’ might be less significant. In Bartle’s original classification schema this group was named ‘killers’, and although we did not use that term we can suspect that many participants knew about it and hesitated to classify themselves as such. It is worth noting that since the players classified themselves, the graph does not reflect their actual play styles: it reflects how they perceive themselves as players, or perhaps even how they wish to be perceived.
Bartle’s (1996) model of play styles is, of course, a simplification of what motivates players; Bartle constructed it as an aggregate model of the responses that players gave to a host of questions. It is likely that most players do not fit into a single category, at least not all of the time. One of the audience members, the famed ARG designer, futurist and academic, Jane McGonigal suggested adding an axis to the plane to see how much deeper a three dimensional version of the Bartle theory could be. Although this is an interesting idea, it is equally compelling to see that the study participants had very little problem in classifying themselves according to the Bartle simple typology. During the experiment, we only used a single board to aggregate the results, so as participants answered, the board filled up. It was suggested that each participant should have had their own sheet, to have a clean view of the two axes of play-style. But with the one board method we used, the participants themselves were able to immediately see the results up to that point and the result when they added themselves to the board. Just like a game, there was an immediate interaction between the player/participant and the system/experiment (with a short tutorial/marketing phase by us).
Our results show that there indeed is something interesting to find out about preferred play styles of people in the game industry. It would be interesting to do the same study but involving all GDC conference-goers to see if these trends endure. Another interesting option is to investigate if there are differences between how players choose to classify themselves and their actual play styles. It could, for example, be interesting to investigate the difference between how players classify themselves and how their friends or colleagues classify them. Another approach might be to investigate how participants might redesign the play style graph based on their own preferred play styles. Although the graph worked in our study, it is not optimal; it is now quite dated and it was developed with one particular game genre in mind. This could be combined with the aforementioned “three-dimensionalization” of play styles.
Reflections
This was a fun experiment and indeed we got to know a lot of people and also the group members and made it very social to be a participant at the GDC. For that reason only one could promote doing real time research, but maybe even more important what could “real” benefits of such research bring us? What data would be interesting to publish from a conference such as the GDC?
For the research area that we drew on (specifically, examination of play styles among varying professions represented at GDC), the study could be viewed as a pilot study of sorts. During the presentation of these preliminary results on the second day of RTR, we had many interesting comments and questions from those in attendance. Though we were under some pressure to enjoy the conference for ourselves while also doing research, the results were indeed interesting and drew eager participation from our audience, which consisted of a great many academics. This was encouraging for both researchers who are thinking about attending future conferences as well as developers with intentions of linking up with the world of academia.
One of the challenges we and other groups most definitely faced was how to approach our subjects, how to let them know what we were doing as fast as possible without taking up too much of their time. Many things are happening at a conference of this stature (ie. GDC), but people were generous enough to give us some of their time and help with our project.
More broadly, RTR could almost be thought of as given Salen and Zimmerman’s (2004) definition of a game as artificial systems in which players engage in conflict, defined by rules and resulting in quantifiable outcomes. We were “players” – our group on the same team, but in competition with other groups (the “conflict”). Our defined rules were our cards. In doing our “research” in “competition” with other groups, the rules and constraints were common. We could choose to follow all or only some of our cards, we could ask assistance from RTR workshop leaders, we could choose our research materials from those made available (pens, stickers, post-it notes, etc.), and we had shared time constraints. All groups had an outcome. Ours was quantifiable and to the best of our memory, so were a few of the others.
RTR could even be said to be a ludic activity - we had a lot of fun during the experiment! One difference between the two is that, in a game, you are in an alternate reality while RTR was “really” real. And while we did not win “a prize” per se, the opportunity to present research results at “THE” GDC could very well be considered a prize in and of itself.
Acknowledgements
We would like to thank Kauthar Tung, Jim Diamond and Jauver Elizondo who were also members of our RTR team and participated in the early phases of this project.
References
Bartle, R. (1996). Hearts, clubs, diamonds, spades: Players who suit MUDs. Journal of MUD Research, 1(1).
Richard Bartle. (2009, August 21). In Wikipedia, the free encyclopedia. Retrieved August 21, 2009, from http://en.wikipedia.org/wiki/Richard_Bartle
Salen, K., & Zimmermann, E. (2004). Rules of Play: Game Design Fundamentals. Cambridge MA: MIT Press
Profession
Post-It Color
# of Respondents.
Business/Management
Green
13 [14]
Audio/Visual
Pink
5
Design
Blue
6 [8]
Production
Orange
25 [23]
Programming
Light Yellow
6 [7]
Other
White
3
Multiple*
* Some participants had several profession titles.
Game Developers’
Descriptions of
their Player
by Carla C. E. Fischer
How do video game developers
describe their ideal players?
How do Video game
developers describe their ideal players?
When faced with the challenge to design a research study, collect and analyze the data, and present it before the end of the 2009 Game Developers Conference (GDC), our group chose to focus on the game developers’ ideal player. Identifying the intended audience is a key element in the game design process. As such, we wondered how game developers would describe the player they have in mind for their current (or most recently completed) game.
Method
During the 2009 GDC, members of our research group approached conference attendees before and after sessions as well as when they were sitting at tables during lunch and asked if they would be willing to complete a small survey, consisting of the following items.
What kind of game are you making?
What is your role in making the game?
List 5 characteristics of the player you have in mind as you create the game.
Please quickly sketch the player you are designing for.
We did not count the number of attendees who declined to complete the survey. Anecdotally, several team members mentioned that they were turned down frequently, reminding them why asking random people to fill out surveys is so difficult.
The other method used to capture the audience characteristics was to ask respondents to draw a quick sketch of their ideal player (four examples of which are illustrated here).
Results
Respondents (n=51) named a total of 247 characteristics. After reviewing the list of characteristics, similar words were edited for consistency (i.e. females became female). Because of the small sample size, the characteristics (discussed below) are not further analyzed by the respondents’ role or the type of game.
Figure 1 is a word cloud of the characteristics, which was generated using wordle.net. The font size of the word provides a visual indication of the frequency by which the characteristic was named. (Color was merely for visual presentation purposes and does not represent any factor in the analysis.) The most frequently named characteristics were social, casual, young, explorer, boy-or-girl, likes, creative, and curious. Social and casual likely represent genres of games. Young and boy-or-girl likely represent audience types. Likes was commonly used in conjunction with another word (e.g. likes puzzles or likes building), and the algo rithms used in wordle.net to generate word clouds breaks the words apart into separate elements. Explorer, creative, and curious are assumed to represent player traits.
The characteristics could also be informally grouped into categories, such as age (generally specified as ranges), gender specifications, personality and lifestyle qualities (e.g. academic, competitive, curious, environmentally-conscious, explorer, social, low technology user, musical, playful, problem-solver, web-savvy), game genre or gamer type (e.g. hardcore, casual, nongamer, gamer, MMO), or cultural descriptions (e.g. English language learner, Caucasian, third-world).
Conclusions
By compiling the characteristics game developers use to describe their target audience, this survey, albeit informal, provides a snapshot of current trends. Rather than focus on the hot genres or technologies, this snapshot illustrates the ideal audience. However, the picture is limited – many of the provided characteristics could fall into several categories, but without contextual information from the respondent, it’s impossible to interpret their exact meaning. Despite this, the concept of quickly capturing five characteristics was relatively easy to implement. Additionally, if performed repeatedly over time, it would likely provide a changing picture of the times.
The other method used to capture the audience characteristics was to ask respondents to draw a quick sketch of their ideal player. While the pictures drawn by respondents were not analyzed, the use of a drawing prompt is worth noting as a method for future researchers to consider. Most respondents’ chose to draw a picture, and they ranged from stick figures to 3D drawings to abstract representations. Respondents’ reactions ranged when asked to draw a picture. Many were hesitant but agreed to “do their best” with a bit of encouragement from the researchers. Others appeared to be perfectly happy to only draw a picture and not fill in the remaining questions. It appeared to be an unexpected and enjoyable part of the survey. It added to the informal feeling of the research and is one that researchers should consider as a method to both break the ice and gather infor mation quickly, assuming the researchers are prepared to do the analysis and coding of the drawings during the conference setting.