Understanding HP Lab’s Twitter Research

Left: The famed HP Labs think tank in Palo Alto.

A few months ago, I spent an entire day with the HP Labs group in Palo Alto, they’re responsible for the R&D and innovation that goes into their thousands of technology products on the market. I was pleased to see this deep dive scientific research on Twitter by Bernardo A. Huberman, Daniel M. Romero and Fang Wu.

You can read the free Social networks that matter: Twitter under the microscope (PDF). Written in an academic style, it’s a bit dense for the casual reader. I write for a business audience and I’ll strip out the most important findings, and add my own insight to what I think matters. As always you’re welcome to chime in the comments.

Understanding HP Lab’s Twitter Research:
If you just need a summary, I wrote this in a way that you can just read the bolded elements to get a sense of the report. I hope this saved you some time.

  • Most users have a smaller inner circle they communicate with: Within a social network, it was found that most only frequently communicate with a small segment of users –even if one has a large community. Makes sense, everyone has an ‘inner circle’. Finding the true network that an individual has (even if they have thousands of “friends”) is what’s really important. Although Scoble solicits imput from thousands of contacts, he leans on a smaller subset of folks to trust above all others.
  • HP Labs Sample Size is @ 6% of the Twittersphere: HP Labs took a random sample set of Twitter users, for a base of number of 309,740 users. According to my social network stats tracking page, Twitter’s total universe is somewhere between 4-5 million (still very small). I’ll value the network on the 5 mil side, so that’s sample size of about 6%, which is pretty healthy.
  • On average, most had 85 followers: They found that the average user has 85 followers in their network, this number seems reasonable when averaged out across the network.
  • On average, most had 80 friends: Most users followed back 80 others, which is close to the actual follower number. Perhaps some weren’t following spam bots, or people that follow everyone. James Governor has been discussing asymmetrical networks, but it appears that on the average, most are symmetrical.
  • Tweet Frequency? About one a day: On average, these users had posted 255 tweets, and since the average users has been around for nearly 7 months, thats about 36 tweets per month, or little bit over one a day.
  • 68%: are active users Social networking stats are almost always flawed, as the vendors don’t disclose how many are truly active. I define active user base as logged in and completed an activity in the last 30 days. Among the 309,740 users only 211,024 posted. It’s unknown if this filtered out spam tweets, although nearly 2/3rds of users have returned (site stickyness. That’s a pretty good return to site rate.
  • Most members have been on twitter nearly 7 months: The research showed that the average person (from first to last post) was active for 206 days. This means that June 2008 (report written in Dec) has become somewhat of a trigger point, perhaps where a growth curve started to point upwards. I noticed an influx of users on April 2008, two months before HPs findings, see comment #579
  • A quarter of tweets (@) are directed at other users: The report showed that Around 25.4% of all posts are directed, by using the “@user” which is responding to others. This could suggest that the other 75% of tweets are updating their network of what users think is interesting or discussing ‘what they are doing’
  • The more followers, the more they tweet –up until a point: Figure 1 indicates frequently in posting the more followers they have, right up until about 500 followers where the frequency starts to level out (if the graph were smoothed). The data around number of friends suggests a similar graph, although there’s no saturation point (see figure 2). I’ll suggest the more connections a user has, the more value they have, and therefore are more active.
  • Despite having large networks, a smaller circle is maintained: For users with a high number of followers, they actually only still communicate with a smaller subset of users. This rule remains constant see figure 4.
  • Where’s the value? within the hidden network: To find out the real value of a twitter user and their network, finding out their true network of folks they communicate with on a regular basis will show their trusted network. Finding out who the Scobles’ communicate with the most will determine will help find out how he is influenced.
  • Business Opportunity for Measurement Vendors
    If you’re a social media measurement company, and can find out the true influence model of who people really trust above all other users by looking at actual “@” behavior and follow behavior, be sure to leave a comment below showing how you can do this. Then, conducting this by topic, will find out the true influencers by market segment within the Twitterpshere.

    Brand Opportunity
    As we know, traditional advertising doesn’t work well in social networks, ‘carpet bombing’ isn’t effective. However, conversational marketing is also costly, as you have to spend great resources on labor to communicate with influencers. Therefore brands who want to be effective with their resources should find out who is an influencer in their market and focus their conversational marketing primarily on them.

    Thanks to the HP labs team who did a great report and really helped to further understanding Twitter better, when you have time, invite me over for lunch, I’m in the area.