Guest Author: Dr. Michael Wu
Lithium Technologies Inc.

After our spectacular LiNC2010 conference we can now resume our discussion of social networks. Last time when I was blogging about this topic I discussed how to find influencers in a community. I described step-by-step how we build the social graph for the conversation network, what data we used, and how we filter the data to ensure content and temporal relevance. Finally, we looked at some real social graph data from Lithosphere using various node metrics that we’ve computed via social network analysis (SNA). Today, I will show you a few more social graphs from different communities to get a sense of the diversity of social graphs out there.

Since I started analyzing social graphs at Lithium, I’ve examine at least 15,000 social graphs across Lithium’s 200 or so clients (over 250 communities) spanning 10 years of history. One thing that I found is that social graphs, no matter how you construct them, are unique; no two communities are even remotely similar. Picking out just a few typical examples to show in a blog is not easy (I can probably write a book just on the different characteristics of social graphs I’ve seen!).

Note: The social graphs in this blog were built using the same procedure that I outlined in my previous blog “Community Influencers Step by Step.” The only differences are:

1. I’ve used a different window length (some longer than 30 days and some shorter) to better show you the structural essence of these social graphs.
2. I also filtered out edges with weak connection strengths to avoid overcrowding the graph because some of these communities are quite large.
3. I have mapped degree centrality (number of connections) to dot size and the PageRank score (authority) to the dot’s color.
4. Finally, since these are real social graphs are from our client base, they will remain anonymous.

Q&A vs. Discussion Communities

Two types of interaction dynamics can be easily identified in a conversation social graph:

1. Question and answer dynamics
2. Discussion dynamics

The Q&A dynamic starts with someone asking a question, then someone (quite probably an expert) answering them. The conversation may go back and forth several rounds before the expert fully answers the question, but the conversation is primarily between two people. Occasionally, other novices may join the conversation or experts may provide an alternative answer, so conversations between more than two people do occur, but infrequently. The experts may also discuss more technical topics among themselves.

This is exactly what we see in Figure 1, which shows the conversation graph of a Q&A community. This graph shows 1366 users co-participated in conversations over a period of 60 days. The experts are those big yellowish dots (denoting high authority) who answer questions from many different novices (the small red dots). The small red dots usually don’t talk to each other, because their questions are usually quite specific, so there are little connections between them. This produces numerous fan-liked structures around the experts. You can find strong connections (brown edges) between several experts who talk to each other frequently. And you can also find several instances where two big yellow dots talk to a small red dot.

The discussion dynamic is very different. A user starts a topic; then many users join in the conversation to voice their opinions. So the conversation graph for a discussion community looks like Figure 2, in which many users co-participated in conversation with each other. This phenomenon shows up as many triangularly linked edges which connect all the users who co-participated in the same topic. I have built this social graph using data from a 15-day period because the edges in a discussion community can be very dense (In fact we compute the edge density for all our communities, and that is how I identify these discussion communities). This graph only has 212 users, but you can clearly see a very dense core of strong connections where each user connects to many others.

Tech Support vs. Marketing Communities

Typical communities will have both Q&A as well as discussion dynamics. There is nothing that prevents a discussion in a support community. Likewise, asking a technical question in a marketing community is not a forbidden activity either. In fact in any given community, there will be a mix of product experts who answer technical questions and discussion leaders who start great conversations (see my video chat with John F Moore, CTO of Swimfish).

For example, Figure 3 shows a typical conversation social graph of a tech support community. This graph has been built with data over a 60-day period and consists of 750 users. You can clearly see the fan-liked structures around the experts as in a Q&A community. You can even see the experts talking to each other. However, there is also a dense core of heavily connected users that mimics a discussion community at the center.

Figure 4 shows a typical conversation social graph of a marketing community. It is built with data from a 60-day window and has 1261 users. Here you can see several clusters of densely connected users. It almost seems like there are three of four sub-communities in this large marketing community. But again, you can definitely see the fan-liked structures around the experts (those high authority yellowish dots) that are scattered throughout the community. There are even two isolated experts, who only talk to the small red dots (one on the upper left and the other on the lower right).

Conclusion

Alright, I hope this blog entry gave you some interesting perspective when looking at social graphs. Was it insightful? I certainly hope so. But where is the art? Well, you’ve been looking at them! Don’t you think these social graphs are beautiful? Even though I’ve already seen at least 15,000 social graphs, the patterns within them are so beautiful that I just never get tired of looking at them. Moreover, since I wrote my own SNA software plus the visualization routines, I can watch the social graph in sequence like a movie. And their growth pattern and temporal evolution are just captivating.

Next time, I will show you two more social graphs and the insights we can obtain via the unconventional graph metrics that we compute. In the meantime, I welcome any kudos, comments, questions, critiques, complaints, etc. 

This article was originally posted on the Lithosphere.

Guest Author: Dr. Michael Wu
Lithium Technologies Inc.

Michael Wu is the Principal Scientist of Analytics at Lithium Technologies Inc. Michael received his Ph.D. from UC Berkeley’s Biophysics graduate program. He is applying similar data-driven methodologies to investigate and understand the complex dynamics within online communities as well as the greater social web. You can follow Michael on Twitter at @Mich8elWu or on LinkedIn at Michael Wu PhD.