Gamification
Data Science

The Right Content at the Right Time: Influence Analytics 3

michaelwu.jpg Michael Wu, Ph.D. is Lithium's Principal Scientist of Analytics, digging into the complex dynamics of social interaction and online communities.

 

He's a regular blogger on the Lithosphere and previously wrote in the Analytic Science blog.

 

You can follow him on Twitter at mich8elwu. 7GY2YSVFQTT9

 


 

In my last two posts I've presented a simple model of social media influence, and I discussed what data can be used to find influencers. Since this article relies on an understanding of the influence model, and I will often make reference to previous blogs, I recommend that you review the articles below before moving forward.

  1. The 6 Factors of Social Media Influence: Influence Analytics 1
  2. Finding the Influencers: Influence Analytics 2

Unlike the influencers, the target we are dealing with here is a much larger and more diverse group, so they are more difficult to maneuver. A simpler approach to ensure that we hit the target is to find influencers who are highly relevant to the targets. And we can achieve this with clever analysis and data manipulation. Since there is much marketing interest in using social graph data, I will use it as an illustrative example and show you how to use graph data properly.

 

Relevance: the Right Information

Last time I already alluded to an example of using social graph data and social network analysis (SNA) to find credible users in a social network. This requires us to trim away the parts of the network that are irrelevant to the domain of interest for the target.

 

Suppose you are trying to find iPhone influencers using the friendship graph on Facebook (FB). Figure 1 below depicts hypothetically such a graph. If you apply SNA on this social graph, you will come to the conclusion that PopGuy is the most influential user, because he has the most friends and he is also well connected to other reputable users. However, this would be totally wrong. Recall from my introductory post (Social Network Analysis 101) that you cannot make any conclusion about any relationship that is not explicitly represented by the edges. People on this friendship social graph are not connected because of their interest in iPhone; they are connected because they are friends. The edge relationship in this social graph is friendship, so you can say absolutely nothing about their influence on iPhone.

Content+Temporal_Relevance_01_resize.jpg

Now, suppose we analyze the data in greater depth, and we have identified every pair of friends within this friendship graph who have talked about iPhones, either on each other’s FB wall or through other FB apps. Now we can build a more relevant social graph, where the edges represent communications about iPhone. I have colored this graph red and overlaid it on top of the friendship graph in Figure 1. The result is shown in Figure 2, and the red social graph in Figure 2 is the content relevant social graph that we should be analyzing, because its edges represent the relevant relationships that are of interest of your target.

Content+Temporal_Relevance_02_resize.jpg

If you trim away all the gray edges, then apply SNA on the remaining red social graph in Figure 2, you will find that Fan08 is the most influential user about iPhone. Since the red edges represent communication about iPhone, we can confidently say that Fan08 is an influencer about iPhone. Although PopGuy still has way more friends than Fan08, PopGuy does not talk to his friends about iPhone; so he is not an iPhone influencer, and the fact that he has lot of friend is irrelevant.

 

Not all social graphs are constructed the same way! The relationship that is captured by the edges in the graph is crucial. Just because you have a social graph, it doesn’t mean that you will get the relevant influencers in the domain of interest. To do that, you must construct a content relevant social graph.

 

Timing: Relevance in Temporal Dimension

Now you may think that the friendship graph on Facebook has 400 Million users and they are all connected, so why do we need to care about relevance? The reason is time! You can seed an iPhone message to an irrelevant influencer that you found via the Facebook friendship graph, and it will spread and eventually reach your target, but it may take a long time. If you seed the same message to a relevant influencer, it will reach your target much quicker.

 

Don’t forget that your target has a “decision timeline,” and there is a small window of time when he can be influenced. Very early on, before the target even enters the marketing funnel, there is essentially no urgency. At this stage, there is little to influence aside from awareness of your brand, because anything you throw at the target will most likely be considered as spam. But very late at the decision timeline when urgency is high, the target probably has already made up his mind. At this stage, it is probably difficult to influence him any further.

 

The golden window is between these two stages, and that is when your influencer can have the strongest effect on your target. If your message reaches your target but misses this golden window because it took too long, you have lost the chance to influence your target.

 

How do we know when a user has entered this golden window? To answer this question, we need to understand the user behavioral norm on social media channels. You probably won’t be surprised to hear that human behavior actually has the tendency to follow Newton’s Law of Inertia; that is, without any external forces, human behavior tends to stay the same. So the golden window is usually right after some actions that the user took. For example, joining a fan group or asking a question in a forum. The key is that we should let activity determine when something is relevant and not just the social graph. Because online relationships and associations we establish usually persist much longer than the time that is relevant.

 

For example, I may join a smart phone discussion group today to find out which phone I want to buy. After two weeks of researching and talking with influencers in the group, I may have decided that a BlackBerry is most suitable for me. I made my purchase, but I never bothered to leave the group, because it’s not costing me anything to stay. Just ask yourself, how often do you unfollow people on Twitter or remove connections on Facebook? My guess is probably not very often. Therefore, to find influencers who are relevant to the target right now, you must also trim away temporally irrelevant relationships and associations that do not have any recent activity when constructing the social graph.

 

Let’s revisit the example using the FB social graph in Figure 1 and 2. Suppose we analyze the FB data further and we were able to collect timestamp data on when was the last time each pair of users had a conversation about iPhone. I’ve color coded the relevant edges (the red edges in Figure 2) using successively lighter shades of red to indicate further back in time. The result is shown in Figure 3.

Content+Temporal_Relevance_03_resize.jpg

Depending on the average length of the golden window for iPhone buyers, we may need to trim away different shades of red edges before we apply SNA on the remaining graph. For example, if most iPhone buyers do research for about 2 months, then we would need to trim away the two lightest pink edges on Figure 3. On the other hand, if iPhone buyers typically do research for 2 weeks, like me, then we would have to trim away all but the brightest red edges on Figure 3 before applying SNA.

 

Although Fan08 has been identified as an iPhone influencer, and he was indeed an influencer, but it was 6 months ago. If you look at Figure 3 carefully, you can see that he only had iPhone conversations with 2 friends within 1 month (indicated by the 2 bright red edges emanating from Fan08). People’s interest changes and those formerly influencers may no longer be influential anymore. Applying the same SNA algorithm on the temporally relevant part of the content relevant social graph (the brightest red edges of the graph in Figure 3) will reveal Expert10 as the true iPhone influencer. Expert10 is the person you want to work with because he has both content relevance and temporal relevance.

 

Conclusion

Once you have constructed a social graph of relationships that have both content relevance and temporal relevance, applying SNA on this graph will give you influencers that you want. These influencers will have the relevant information that your targets want. These influencers will also be the ones who can deliver the information to your targets before the expiry of their golden windows.

 

In my next article, which is the last in this series, I will look at the final step of bringing the influencers and the target together. Meanwhile, any questions, comments, critiques, or kudos are welcome. Drop me a line here so we can carry the conversation further.

 

 

 

Related Blogs

5 Comments

Thanks Michael as I understand this what you're saying is that using SNA to find the right influencers to influence target users' purchasing decisions (bearing in mind the value of the targets' first online activity around a product as the indicator when to start the persusion process) relies on the finding the right type of influencer - which is not the one with the most friends as the connection, or with the most discussions around a product, but the one with the most recent discussions about the product: relationship + product discussion + timely = best chance of success. Reminds me of my own blog post about how communities managers can play a role in this optimisation leading to revenue!

Nice article Michael.  Totally agreed that there is not "one socialgraph".  In fact, for different applications one key step is the construction of the appropriate social graph. 

 

One question for you:  do you make the distinction between influential and connected?  There are some vendors who say that the most influential are not the most connected.  It comes down a little to defining which graph metrics you are using for influence, is it degree centrality, eigenvector centrality, pagerank (using replies and posts and directed links) or some other?

 

Sincerely,

Nick

Sonamine, LLC

www.sonamine.com

Gamification
Data Science

Hello Stuart,

 

Thank you for summarizing the essence of this blog entry so concisely in your comment... Smiley Happy  What you said is perfect:

 

"The right type of influencer - which is not the one with the most friends as the connection, or with the most discussions around a product, but the one with the most recent discussions about the product: relationship + product discussion + timely = best chance of success."

 

However, one must keep in mind that there are 6 factors of social media influence. This blog entry only covers 2 of the factors. And so far, we have only covered 4 out of the 6. The last 2 factors will be discuss next week. So stay tuned!

 

Gamification
Data Science

Hello Nick,

 

Thanks for the comment. From your response, I get the sense that you have worked with social network and social graph data. What you said is very true: most of the hard work is to construct the appropriate graph for the solution you need. Using the wrong graph (as I’ve shown in this blog post) can lead to erroneous results.

 

We DO make the distinction between connected and influenced. But the distinction is sometimes confused by the language we used. Connected is a graph property, but it also has a social connotation to it, yet influenced is a social/marketing concept. It all depends on what is relationship is being represented by the edges (see my intro post “Social Network Analysis 101”).

 

For example, if the edges simply represent friendship, then connected on the graph has little to do with influence. If the edges represent conversations about a particular product, then connected on the graph would mean a significant potential for influence. If the edges represents a purchase decision or action, then connected on the graph would be equivalent to true influence.

 

I agree with “some vendors” that the most influential is definitely not always the most well-connected individual. There are 2 levels to look at this. First if you don’t have the relevant graph, then definitely the most well-connected will not be the most influential (you've seen an example in this blog entry).

 

But even if you have built the appropriate and relevant graph, the number of connection still doesn’t always imply influence. Connectedness, or number of connection, is just one of many social graph metrics, namely, degree centrality. There are many ways that an individual can be influential or important in a graph. As of now, we compute 10 different social graph metrics for all nodes on the graph. Five of these (which include all the ones you mentioned) are described in another introductory post “Are all Influencers Created Equal?” Depending on the different marketing constraints and goal, you will need help from different types of influencers.

 

I have already written a post that will briefly describe the other 5 node metrics we compute. It will be published in couple of weeks.

 

I hope this addresses your questions. Thanks again for asking.

 

 

I read this now, and think Google Zmot :-)

Lithys

Lithy Awards 2017

The winners in digital CX have been crowned!

See the winners!!