The Big List Blog

| All |

The science of social — Live from the Measurement Pre-Conference

Coverage of this session by Bridgette Cude of SocialMedia.org. Connect with her by following her on Twitter.

11:55 — SocialMedia.org’s Kurt Vanderah introduces Lithium Technologies’ Chief Scientist, Michael Wu.

11:56 — Michael: Our marketing team sent out a short blurb on the social customers, so I’m actually going to talk more about the social metrics.

11:57 — There are many social metrics out there. But when you talk about ROI, people want to know what happened at the business levels.

11:58 — These are a completely different level, so how do you bridge that gap. Social analytics tries to do this, and often fails, but they try. For example: People don’t measure sentiment. All of these are computed and derived from the raw data.

11:59 — How do you go from the operational level to the social CRM? There are two types of models: the correlation model and the mechanism model. We all know that correlation is not causation — so how do you prove it. What you need is experimental manipulation. With data only, you cannot prove it. “You cannot prove smoking causes cancer with data alone.”

12:00 — Here’s an example of how to use the mechanism model: There’s a lot of data generated on social media, but no one has it, they have social media activity. They infer when social media influences purchases. What they have is a “black box model.”

12:01 — Through computation, you get an influence score. Anyone can come up with this model. But it’s not actually validated, and the accuracy is unknown. How do you know how accurate a Klout score is, for example? There’s not feedback mechanism. At best, it’s sparsely validated.

12:02 — To actually quantify influence, we need a more sophisticated math tool: A social network is a collection of entities and the connection between them. They’re called nodes and vertices in SNA language. Relationships are called edges. A social graph is the diagram of these nodes and edges.

12:03 — All of these people may be on the same network, but they’re on all different graphs. For example, I might be on LinkedIn, Facebook, and Twitter and have different relationships on each. Michael explains a graph showing relationships in a simple social network between co-workers, drinking buddies, bosses, etc.

12:04 — Facebook clasps all of these people together and calls them “Friends.”

12:05 — Social analysis shows the social graph, but shows which relationships you care about.

12:06 — You have to interpret this intelligence to understand who these people reach.

12:07 — When you’re reading a social graph, you have to figure out what the edge represents. But don’t make inferences about these. You can’t infer influence. Just because I am your friend, doesn’t mean I influence you. I just have the capacity to.

12:08 — What classifies influence?

1. Domain credibility
2. Bandwidth (The influencer’s ability to transfer his thoughts on the channel)
3. Relevance (A car salesperson cannot influence your opinion on buying a camera)
4. Timing
5. Alignment (You have to be on the same channel)
6. Confidence (How much does the target trust the influencer — not the same thing as domain credibility)

12:09 — If you’re missing any one of these factors, you could miss the potential influencer.

12:10 — To find the potential influencers: Social media marketers are very focused on the “high bandwidth users,” or the “loudmouths.” Reputation: Self-declared and potential people to influence.

12:11 — The popular guy is not necessarily the influential guy. He may have tons of irrelevant relationships. You have to find the relevant relationship. Then, the loudmouth might not necessarily have the right timing.

12:12 — Finding influencers is intelligent filtering. The easiest way to do it: Figure out if they’re on the right channel. Timing: If the relevant time is now, how far back to you go? It’s different for different products. Then you look at bandwidth, credibility, and then you’re left with people who may have relevant influence.

12:13 — Now that we have influence graph, what metric to we compute? For example, Google uses PageRank to find which page is most authoritative — it’s a very similar problem with finding the right influencers.

12:14 — Eigenvector centrality is similar to Google’s PageRank

12:15 — “The ability to quantify something has nothing to do with whether it actually matters.” -Albert Einstein

12:16 — Michael shares some results from his JMR article: Influence does matter the increase from influencers is about 49%. — Out of time –

Q & A:

Q: Are there dynamics that are BtoC vs BtoB specific?

A: Michael: The timing is actually much longer for BtoB companies than BtoC.

Q: Could this possibly be a scoring system for places like Klout?

A: Michael: These platforms have the potential, but they ‘re not. At Lithium we’re not in the business of measuring influence, but rather cultivating it.

Q: Can you help me understand the difference between Confidence and Credibility?

A: Michael: Confidence, or trust, is an attribute of the audience. Whether I have credibility or not is an attribute of me. They are related, but they’re not the same thing.

Q: When message penetration is the end goal, does the timing window broaden? Linking behavior: is there a credibility addition to this? Does linking to third party sources help build it?

A: Michael: The first question: When you try to drive awareness, the timing is actually shorter. For example, how long do you have to see something to not forget it? Second question: It does, but ultimately, you should not do it for the sake of linking, then people will find out and basically you will lose trust. When you don’t have all six factors, you break the chain.

Q: Are you able to reveal the length of their influence?

A: Michael: This is the study that I did not get to go into detail about. It is hard to quantify the value of influence because it is so pervasive. Who has influenced me? The method we use is total impact upon removal. We remove someone from the social network and see how the value is affected by the loss.

December 9, 2013 0 comments

Comments are closed.