Thursday, March 24, 2022

Facebook Algorithms

 

The extended ERJ family may have been bitten by Facebook algorithms.

Belladonna informed me that there was a major kerfuffle going on between one of my brothers and one of my nephews.

What dragged me into it was that the nephew claimed I had done something that had not happened. Or so Bella informed me.

Skipping over all the garbage in the middle, the kerfuffle transpired 15 months ago and my nephew somehow omitted the fact (15 months ago)  that the activity I shared with him happened at a public park and not on Mom's property.

Mulling it over in my mind, it occurred to me that Facebook exists to sell "eyeballs". Most of their revenue comes from advertising. They undoubtedly resurrect threads on a regular basis based on various metrics.

One metric would be Reaction Score. How many viewers, as a percentage, comment on the thread. Another metric might be how often viewers open a given thread and review the latest comments. Stickiness, how long do viewers remain interested in a given thread.

This is not necessarily a bad thing. Threads associated with your high school winning a state title in basketball, for instance. Facebook might notice the bright-spot in their significant-metrics index and resurrect those threads on the one-year anniversary of when those threads peaked. Most people who viewed or commented would be pleased.

The dark-side is that those same metrics will plow the scar-tissue back to the surface for petty drama.

The take-home is that anybody who still plugs into Facebook should always look at the dates of the comments that get put in front of their eyes. We should also coach the Belladonnas of the world to do the same.

There is enough real pain heading our way. We do not need the distractions of phantom targets painted by social media algorithms.

Iterative methods for eigenvector searches

For those of you with a pathological, mathematical bent: The social-media algorithms are very similar to the iterative methods used to find eigenvectors. Multiply a matrix by a vector and you get a new rotated-and-magnified vector. Normalize the new vector and cross the matrix by the new, normalized vector.

The vector rotates into the dominant eigenvector and the other eigenvectors can be found by taking the difference between the last two iterations and enforcing orthoganality with previously found eigenvectors.

Marketing clinics are similar. Take the comments and observations from an earlier focus group and feed them into the new focus group.

As Bob Little once said "Actually, it is rocket-science. But don't panic. We are engineers."

2 comments:

  1. F#@kbook:
    They make some money on advertising, but the big money is selling your personal information;
    They even track you when you are not using F#!kbook.

    ReplyDelete
  2. When I get some time I'll Google all of those words you used in that last chapter and try to figure out what you were talking about.---ken

    ReplyDelete

Readers who are willing to comment make this a better blog. Civil dialog is a valuable thing.