Skip to main content

How much does your voice count on Twitter?



The general Pagerank method discussed in class is extremely versatile: as long as we have a connected network, we can approximate the relative importance of the member nodes through the number of links or connections to them. An article discusses a possible application of this model on social media services to observe the process by which personal opinions form through online interactions. The graph, in this context would be made of the following two components: human participants as nodes, and endorsements (often in forms of “shares”, “retweets”, etc) as directed edges. For example, if A endorses B in some context, there would be an edge A->B.  After the algorithm runs by evaluating relative importance of each endorser, it outputs a numerical value representing the amount of endorsements B gets–a metric we could interpret as the relative importance of B’s opinions within that social platform. The method, if it works, can identify users whose view are endorsed more than the others’. We could call this set of individuals influential to society. This type of opinion formation does have convincing examples outside of the web, however informal they may be. Talk show hosts, ranging from Bill Maher to Jimmy Fallon, gains a social influence because of the fan base that “follows” them. They are often seen using this power in their public appearances to address current political issues. Also, with a very similar logic, we often assert that scholarly articles endorsed by influential papers are also often influential.


Three obvious limitations to this model, however, questions the validity of this approach. First, this type of analysis requires that each online member’s posts are about their opinions, and there is a well-defined user action that can be interpreted as endorsing someone’s opinions. As a trivial example, an online humor account can hardly be considered politically influential and respected simply because the funny pictures are well-received among the users. The opinion rank of this account would be extremely overestimated. Second, the SNS platform should have a well-defined user action that can be interpreted as opinion endorsement. The example of Facebook-ian reactions may apply here. Twitter’s retweets reasonably represent a direct endorsement of an opinion since the exact content of the endorsed appears on the page corresponding to the endorser. However, reactions on Facebooks can be ambiguous: an “angry” react could imply either an empathy towards the original poster’s disappointment or an actual disappointment towards the original poster. In these examples, setting up the inter-node links is trickier. Finally, one of the basic assumptions in the model above is unfounded. The original Pagerank model with websites deal with chain of links from one website to another, and this chain by itself also denotes a valid link. However, the existence of a chain of endorsements may not imply the possibility of a direct endorsement. This is because human beings are multi-faceted and so are opinions. In this case, counting the aggregate endorsement for the opinion rank is unjustified. The model is at jeopardy. Perhaps the model can become more accurate by only counting small chains of endorsements such that the probability of direct endorsement is much higher.


Leave a Reply

Blogging Calendar

October 2017
« Sep   Nov »