Home > Misc > Follower Count: Meaningless, Follower Temporal Interest Graph: Priceless

Follower Count: Meaningless, Follower Temporal Interest Graph: Priceless

December 5th, 2009


I have been thinking about some of the current approaches to measurement of social network strength, and realized that there were a few specific things that seemed to be missing in these approaches. In this post I review why the complexity of social network behavior far exceeds a single simple metric, and propose a multi-dimensional model to describe social network strength and utility towards a selected objective.

Power MeterFollower Count: The Simple Metric That Isn’t

Follower counts are generally mistrusted for validity, but still used due to simplicity. Follower counts are, in some ways, a somewhat bizarre way to measure social network strength. General follower count is a “feel good number” that in the end means little in terms of viability of achieving a particular goal.

I keep a few examples in hand to debunk the concept that high follower counts automatically create generic utility. Here is one of my favorites. Sockington (http://twitter.com/sockington) is a Twitter account owned by Jason Scott’s cat (well… take a look and you’ll see what I mean). At the time of this writing, @Sockington is closing in on 1.5M (yes million) followers. If you want to send a message to 1.5M people to entice them to, for instance, buy life insurance, would the reach of Sockington’s network be substantial? Maybe in terms of the old school “impression count,” but that’s probably about all. In the process, if you were persistent about it, you’d also see a pretty high abandonment rate, and I suspect that most people who follow Sockington would ignore such a push message as advertising spam. Certainly if the account and Sockington’s network were used more than a very few times in pursuit of this objective, it would likely fragment and be disrupted as a result. Such a message is not why they follow Sockington, and any other sort of message is both out of character and not their impression of why they follow him. (I’ll resist getting in to the psychology and implications of 1.5M people following a fake cat’s Tweets for now!)

Clearly the raw count of followers isn’t what we’re after. Some have proposed measures of engagement, using reply counts and conversation rates on Twitter, or comment posting rates on blogs.

Simple Event Count-Based Aggregate Network Strength Also Misses The Target

Dart arrows missing targetUnfortunately, measurement of social network strength also defies a single numeric count of followers or responders or even conversational length. Various client and personal experiences have led me to develop a concept I’ll call a Follower Interest Graph. A Follower Interest Graph is a multi-dimensional graph of a social network that maps several factors that define the strength of the network for a particular utility and goal. Before talking about what a Follower Interest Graph might look like, let’s look at some assumptions and realities of social network behavior.

Social Network Topologies Are Fluid Across Topic And Time

Flowing river
Social networks are not monolithic across topic areas, or stable in size and topology over time. Certain posts, comments and topics draw much more involvement than others, in the same social network. How can this be if there is a single metric of “engagement” of followers and participants?

Topics that engage a community are time dependent. The discussions and interest levels follow trends of involvement. A topic which might lead to many discussions and many clicks through to a resource at one point in time might draw a tiny fraction of this engagement at a different point in time.

How can a single value describe network strength if the response and involvement rate fluctuates by an order of magnitude, or more, based on the specifics of the item published or the time at which it is published?

Complexity Of Social Networks Far Exceeds A Single Descriptor

It seems to me that the reality of social network topology includes at least the following concepts (there are more, but perhaps this is a workable starting point):

Social Networks Are Meshes and Change Dynamically

Social networks are strongly meshed, and change over time, sometimes quickly. Follower interest levels change over time, and they may disconnect from the mesh for any number of reasons. While still connected, they may loose contact with the details of the network if they increase their network size past their ability to handle the volume of updates.

Connection Localization Matters

I assert (in the absence of having found any statistically valid research) that the median link distance from a person to the bulk of their active connections has a strong correlation to their influence level on those connections. If you compare the influence strength of two people with equivalent counts in their total connection graph (direct and indirect through connectors), the one with more connections at fewer hops from themselves in the graph has a stronger network.

Why does localization of connections matter? Distant connections are more fragile and carry less signal to noise for the receiver, as you move more links from the center point of the graph. The likelihood that a person five hops away will continue to be actively interested in the center point person’s commentary without directly following that person is suspect in my mind. This “loosely coupled mild interest” effect, more than two hops from a source, is evident in my and other social networks.

A behavioral change by a person in the intervening link system could cause any number of people farther away to disconnect, independent of the quality of the source person’s material, or the target’s interest level. This is one example of the lowered stability factor that is  brought on by “distant interest connections.”

Follower Interests Change Over Time

People are not machines. Interests appear and disappear spontaneously. Network connections appear and disappear spontaneously. Prioritization of interests will shift over time, based on personal situation, impact of broader global, societal or political issues, or boredom with a topic.

Participants in a network often participate in a person’s network due to interest in some part of what the issuer has to say. They may have no interest in a significant amount of the material generated by the issuer – but some part of it is of high value to the follower. Items below the interest threshold are discarded, deleted or ignored. If the receiver’s “personal noise level” gets too high, a disconnect from the network will be triggered.

This behavior illustrates that participation in a network does not guarantee nor indicate which of the network’s topics are of interest to the people participating. In effect, they follow a particular network opportunistically – to get what they want from it, while treating the rest as noise that might as well be push advertising (and sometimes is!)

Intent Influences Reaction, And Thus Network Strength

A mantra of social media has been “pull, don’t push”. If a network is based on provision of information solely or social interaction, with no commercial connection visible at any level, this sets expectations for the participants. If the intent of a particular message that “uses” the network is obviously different than the core expectations of the network participants, it may be fully ignored (looks like advertising) or worse may cause damage to the network (either in terms of defections or trust reduction.)

This leads to another observation: The strength of a network, or it’s effectiveness towards any particular goal, depends on the conversation and messaging towards that goal being consistent in tone and intent to the typical conversation and intent of the network. Attempting to “spin” a social network for intents quite different or even contrary to its normal use is likely to backfire. In misusing a social network, in a fashion counter to the participants’ expectations, you will likely disrupt or in the worst case destroy the network. Thus, although large social networks may be attractive to marketers, if they are not built with their eventual use cases in mind, they may have a very limited lifespan and represent an expensive investment with a limited lifetime of usefulness.

Follower Temporal Interest Graph: A Reality-Based Solution?

3D numbered cubesA Follower Interest Graph maps a social network using more factors than simple counts and response or engagement rates. It combines the real dynamic factors of network influence and viability into a single model:

Social Connection Count

The Social Connection Count is the total count of connectors at some linkage distance from the person in question.  What is the effective radius of a person’s network – not quite known, but mabye 3-4 hops, if that? My hunch is that connections past 2 hops should be de-rated substantially, but this is not substantiated by more than a gut feel at this point.

Social Hyper-Connector Localization

A visualization and metric possibly represented as a heatmap that illustrates the density of relevant topically interested followers at different linkage distances from the person in question. The numeric representation of this might be the median number of hops from the person of interest that encompass 80% of the connections (I have not really thought through where the cut-off would be for this – at this point this is a conceptual model only)

Temporal Sensitivity

As discussed previously the strength of the network for a particular topic will fluctuate pretty dramatically over time. Perhaps this is a range of values illustrating minimum engagement in a down cycle versus maximum engagement at the peak of a topic trend.

Topical Focus

Given that the network topology and count will be dramatically different across topics, it seems that any temporal interest graph would be valid only for a particular topic or topic area.


Intent of the publisher in question and of the participants may match closely, peripherally and may shift over time. Which intent areas are most favorable for different parts of the network, and how does this match with the marketing goal-driven usage.

What Does A Follower Temporal Interest Graph Look Like?

At this point, I am more interested in feedback on the core concepts and approach for this metric, and less concerned about visualization. If it’s a good metric we’ll find ways to visualize it!

I suspect that a three-dimensional representation would capture most of the meaning of this complex metric. One approach could present a three dimensional, color-coded heat map of connector interest in a particular topic on the X/Y plane (showing the density of topical interest at various numbers of link hops from the center point). The height (Z axis) representing the changes of interest over time. I’ve done some napkin sketches, but they are far from ready for prime time, and hoped to get some conceptual feedback while trying to get them in to better shape for discussion.

Looking For Feedback!

I figured it would be worthwhile to put this idea on the table before spending a lot of time refining it, and I hope that it is at least organized well enough to hold together. I am interested in any and all feedback on this concept, to see if perhaps this could form the basis of a workable model for social network utility.

Thanks for reading! (This has been a long post… even after some editing, and I appreciate your interest in walking it through to the end. )

Categories: Misc Tags: ,
  1. December 6th, 2009 at 00:37 | #1

    You might find this classic 1983 paper on “The Strength of Weak Ties: A Network Theory Revisitied” to be interesting.

    http://tinyurl.com/strong-weak-ties (.pdf direct download)

    It has over 2000 literature citations to date. The original paper was in 1973 I think, and is a classic of sociology and social network theory. The strong ties vs weak ties analysis is fascinating, and I think it’s connected to what you are trying to do. I suppose it partly depends on how you define “strong network” and what purpose you see it as serving. What do you think?

    — Chimera Cosmos

  2. December 6th, 2009 at 08:32 | #2

    @Liz Dorland

    Thanks Liz. I’ve read that paper, and re-read it based on your mention. I’m after something a bit different here. Perhaps talking about the strength of a network seems confusing with the strength of individual links.

    A central point that I am exploring is that the utility of a network in accomplishing a goal – getting its members to take action of some sort as opposed to merely passing information – is based on a dynamic set of parameters that are more than just an abstract shape of connections.

    I believe that the utility of a network in motivating action is highly dependent on a match of the topic of the particular item to the interest map of the participants and a temporal component, such that the effective geometry, distribution and ultimate effectiveness of achieving an action goal changes dynamically.

    This is contrary to the monolithic view of network topology discussed when people talk about the size of a social network, wherein the implication is that a social network, once the connections are created, they some fixed utility across topic and time. I believe that having a network of social connections is a baseline, a possibility, and that whether or not all of that network is energized by a particular topic or call to action is a fluid, dynamic question related to the match of the topic and intent to the people in the network.

    I’ll keep tinikering with the semantics to clarify this better. To me, it seems that almost all of what we call large social networks are relatively weak ties, however in action it still seems that these weak ties are more dynamically unstable the more hops you get from the “source”, or the person at the core who may attempt to be a thought leader or motivator for a network-level action.

    I would argue that the likelihood of a particular topic or intent to jump from one cluster (strongly linked group) to another via a weak tie is more affected by these interest and temporal effects than the likelihood that a topic will travel within a strongly linked cluster. This effect may be part of why attempts to generate a viral message have had very limited success rates.

    By dynamically unstable I mean that if you mapped the spread of a specific topic, message or intent, the likelihood that it would reach and be relayed by a hyper-connector far from the source is based on a number of factors linked to intent and timing, and that the key linkages that lead to that remote hyper-connector may well have disconnected over time. In addition, the remote hyper-connector, in today’s networks, follow so many people (often in the 10’s of thousands) that their actual follower list is a minor fraction of their apparent network size.

    As mentioned in the paper: “Weak ties provide people with access to information and resources beyond those available in their own social circle; but strong ties have greater motivation to be of assistance and are typically more easily available. I believe that these
    two facts do much to explain when strong ties play their unique role.”

    The likelihood that a particular motivator, topic or call to action will create results seems to be highly topic and temporally dependent – connected to whether it matches the owners of the weak ties – in determining whether they will pay attention to the message and find it compelling enough to generate results (even if that result is to “pass the message along”.)

    Thanks again!

  1. No trackbacks yet.
Comments are closed.