We describe the approach and initial results of a study looking at how technology communities contribute to learning for software development professionals and how, in particular, they make extensive use of Twitter as a means of transmitting and eliciting knowledge within and between groups. Collected data from Java User Groups (JUGs) on Twitter around the world shows us, on one hand, how a tweet communication pattern shows evidence of Knowledge Transfer and, on the other hand, how clustering of JUGs around geography, language and culture can function as a barrier to the diffusion of Java knowledge.
1. COMMUNICATION BETWEEN JUGs
Java is a programming language with an estimated 9 million professional developers  worldwide, many of who gather around JUGs in informal associations for technical discussion; these are communities where people feel relaxed and able to exchange knowledge with their peers in a free and informal way. These communities’ members are all in a social dual mode, collaboratively teaching and learning at the same time, in a process defined by Dib (1988)  as a type of Non-formal education “…any activity or structured educational program organized outside of the formal system and which does not lead to certification from that system”.
Today’s technical job requirements change so fast that when people return from an external, formal training, their newly acquired knowledge is already in danger of obsolescence. Professionals looking for instant support to resolve their daily problems find that 3% of the questions posed by developers to the online forums of one such user group received their first useful suggestion within 1 minute .
According to Oliveira (2003 p.IX)  learning opportunities in JUGs happen through face-to-face periodic technical lectures, discussion lists, personal or community blogs, common shared jokes, articles in news aggregators, chats, e-mails exchanged between members, and videos on YouTube. Originally paper-based, but now almost entirely electronic; more and more Java groups are broadcasting news and announcements through social networks like LinkedIn, Facebook and Twitter.
When examining the visible evidence of knowledge sharing between JUGs, we observed that use of Twitter by JUG members was very widespread and integrated into their conversations, possibly because it is no-cost and globally available.
2. JUGs STRUCTURE AND PARTICIPATION
Through personal contacts and general Internet search tools we identified 316 JUGs worldwide, in 88 countries and all continents, and found them basically structured in four layers: Core Group, Active Members, Peripheral Participation, and Outside Connections. Our model for the structure of a JUG is based in the three actors proposed by the Degrees of Community Participation model (2002, p57)  to which we have added the group of Outside Connections, as shown in Figure 1.
The Core Group are the community leaders involved in the daily management and support of the activities of the group, either social or technical. Active Members are people that at least post a message or participate in a face-to-face meeting once a year. Peripheral Participation are the Lurkers, the vast majority of the community, they don’t visibly participate in the activities of the group and represent around 80% of the community. Outside Connections are invited speakers sometimes from other JUGs, or industry specialists employed by the technology owner as Evangelists. They are not members, but they are a very important component of the community. Takhteyev (2012) called them the “alpha-geeks”, the “powerful globe-trotting information brokers”, who are the itinerant practitioners and whose goal is to circulate worldwide, spreading and synchronizing situated knowledge for a common understanding (p.13, 27) . In this process of cross-pollination, they take the role of “Bees” sharing knowledge into the community and, making the bridge between different groups.
An account named @JUG_research was created on Twitter, and a list of all identified JUGs in the world were added in order to facilitate the process of data collection and analysis. Observing the flow of information, through Social Network Analysis on the data collected, we could analyze patterns of how information moves within and between JUGs, measure the quantity of tweets generated, observe community clusters and, from the timestamps of posts, gain an idea of the temporal flow of the information. From the collected data, 3 random samples of 100 messages were then coded by hand, using an approach based on Thematic Analysis . All Twitter “Promoted” messages (advertisements) were discarded.
The proposed Participation model described above was validated by observation of the messages (tweets) exchanged between JUGs and suggest that the communication patterns are evidences of transferred knowledge and an immediate precursor to learning. From 316 JUGs worldwide, we have identified 242 that have Twitter accounts (77%), so the use of Twitter by these groups is clearly widespread and seen by members as part of the communication activities of the JUG. We analyzed (using NodeXL ) a total of 167,659 tweets collected from these JUGs, covering the period from May 31, 2008 until July 29, 2015. Based on the number of Followers of each Twitter JUG account, we estimate the JUG population as 894,000 members.
All four types of actors were identified: Leader, Member and Outside Connections, who were responsible for generating all the analysed tweets, and (by exclusion) the Lurkers, followers of the JUG who never tweeted. 60% of the tweets analysed contained “new” information and 40% were retweets.
Figure 2 – On the left the graph shows JUG clusters and exchanged tweets between them. On the right is a chart showing the seven predominant languages and the number and percentage of JUGs.
Nearly half of the tweets identified as representing information transfer (49%) are Invitation tweets, to a meeting (32%), to read something (15%, which we will comment below) or to download some information (2%), as shown in Table 1. Other categories found are: pure advertisement without any technical information content (20%), tweets written in languages other than English, Portuguese, French or Spanish (16%), making a Judgment (positive or negative) about some other tweet (9%), administrative content of the JUG (3%), Job offers (1%), JUG Leaders asking members to suggest lecture topics (1%), and Leaders asking for some kind of feedback about a past lecture (1%).[*]
As well as being followed by their members, JUGs follows other JUGs, as the above figure 2 (on the left) shows. For instance vJUG of London, is followed by 35 other JUGs around the world. Figure 2 (on the right) shows 8 JUG clusters where geography, language and culture is the determinant factor of agglomeration. From the top, reading clockwise, the first cluster is the Japanese community, then East Europeans, a group of two Indian and Pakistani communities, the Brazilians (at the seven o’clock position), a group of Spanish Latin American JUGs, the Germans and finally the French JUGs at the eleven o’clock position. Scattered in the middle of the graph can be seen JUGs that only tweet to their own community and are not connected to the global network.
[*] A limitation of this work was that, besides the authors speak English, French, Portuguese and Spanish, which allowed us to analyze 84% of all Tweets in the sample, the remaining 16% were classified as Other Languages. By observation usually Japanese and German.
5. CONCLUSIONS AND FURTHER WORK
Tweets exchanged within and between JUGs evidence a pattern of information and knowledge transfer within and between communities which, although rooted in the real world, nevertheless interact in the virtual space offered by Twitter. However, we also observe that geography, language and culture of JUGs act as a barrier to global knowledge diffusion. The planned next step of this research is to look who are the people behind each JUG, the ones that take the decision to tweet or retweet messages to their communities. What are their motivations toward collaborative learning? To answer this question, we plan to use their narratives, the unstructured and qualitative data of their tweets.