Andreas Weigend | Social Data Revolution | Fall 2012
School of Information | University of California at Berkeley | INFO 290A-3

Class 7: October 22, 2012 (4:15-5:45pm)


Responsible for initial page (up by 10pm on Thursday after class):
  • Brandon King (kingb ~~a~~ berkeley.edu)
  • Omar Rehmane (orehmane ~a~ berkeley.edu)
  • eunkwang joo

Class materials:

Influence

What is Influence?

  • influence - the capacity to have an effect on the character, development, or behavior of someone or something, or the effect itself (Oxford English Dictionary)

Part of the goal of todays class is to figure out what is influence, and is it even possible to measure it and some it up in a score, like klout.com is doing.

Understanding Influence - Giving it a Score

  • Gandhi - Known for non-violent civil disobedience
  • Hitler - Known for the Holocaust and World War II

The question now, is which of the two was more influential? Is it measure by the number of people who where involved? Is it measured by how positive or negative it was?

Factors of Influence

  • Quantity - How much influence, as in how many people where influenced?
  • Strength - Of the people influenced, how strong of influence was it (making someone laugh vs. got someone to buy an expensive product)
  • Intension - Did the ultimate influence of a person result in the intended influence (ex: I wanted my friends to vote Yes on Proposition ##, but instead the conversation lead to most of them voting no.)
  • Hidden/Subconscious - The ways in which we influence people without realizing we are doing it.
  • Trust - Does the influence depend on trust level? If so, does that effect Quantity and Strength? (i.e. A doctor says something about health vs. sales person)


Defining the Influence Score Black Box


Questions:


Can influence be summarized into a single score? Is Influence:

  • 1 Dimensional
  • Multi-dimensional

Is summarizing Influence beneficial or harmful?

  • Could creating a score be like the token vs. money experiment in the following TED talk? In other words, does creating an influence score disconnect us enough from the true meaning that we loose sight if what it actually means?:

Is there a ground truth? Can we measure it?

  • Assuming influence is multi-dimensional (Think of it like IQ Score is an average of many subtest scores), then a single score would be an average.
  • How much influence is online vs. offline?
  • Are scores based on interactions on Twitter, Facebook, and Google+ able to capture someones true influence level? What about someone who does not use online social networks?
  • What is the difference between Influence and Perceived Influence? Which matters more? Which are we able to measure?
  • What golden standard can we use to measure correctness of an Influence score?
    • What are the consequences of getting it wrong? Especially when we think we are right:

CEO of Klout:


Joe Fernandez, CEO of Klout, had some interesting things to say on the concept of influence. Klout is a website that aggregates online data to assign an influence score to each user.

Influence is not that different from emotions like love and anger, it creates a response based on perception. We've come closer to being able to quantify it, but it's not an exact science or a direct calculation Ultimately, it's down to the perception of influence rather than any quantifiable attribute. A lot of the data must be abstracted - billions of pieces of hard data are collected and processed each day, but they also have to use signals in order to determine a score for each user.

The issue with attempting to quantify influence is that it's difficult to decide on a ground truth. In general, due to influence being a perceived trait, we rely on passing a "sniff test," people thinking the numbers are reasonable. If people change their level of trust, something needs to be changed. (i,e, bloggers having a greater influence than President Obama) In addition, it only takes into account online data, and is thus limited in scope.

As to the upshot of assigning influence scores to people, it is a quick way to determine how effective someone is in terms of social data (collecting likes, followers, etc.) and how well they leverage their abilities. There are also downsides, however. People take it too seriously, much like IQ scores, where people used them for things they didn't actually measure, like job offers. Scores can also change rather quickly, perhaps faster than influence can actually reasonably change. Language barriers can interfere with data collection as well, but most of all, the influence that something has tells us nothing about its meaning or positive/negative effect.

A perfect way to calculate influence is somewhat difficult to come up with, as there will always be some averaging of data, which flattens out important details. But Klout's data integrity is rather good - users cannot buy higher scores, their algorithm deals with that.

The idea of measuring influence, though a difficult and somewhat hazy one, is definitely here to stay, and might even be considered a natural development in the Social Data Revolution.