Facebook has recently begun a trial of its new "graph search engine." This should scare the living daily lights out of anyone, for the reasons I'll explain below.
(...)
For those of you who are mathematically and statistically inclined, this is what a "decision tree" looks like. If I assign probabilities to each vertex for a possible outcome, I can then calculate the probability of eventually reaching the final right hand vertex. If I don’t "know" the probabilities, I can use Markov Chains to figure out what the right distribution is. If I’ve confused you so far with the math mumbo jumbo, stay with me as I give you an example of the usefulness of this approach. Say I want to know what characteristics "terrorists" share in common, without really knowing. So, I might come up with a bunch of things that I think terrorists share in common with each other: such as "likes" Osama bin Laden, writes about (or is a member of an Islamic group), reads various patriot web sites, etc., etc. I can then seed the model and compare it against the cached web and facebook comments of all of my "known" bad guys and have the model converge on a list of the common descriptors. So, in my mini network in figure 1, vertex 1 (the far left one) is you. The far right vertex represents the "known terrorist" convergence. Progressing frorm you, there are two possible attributes (remember, this is simplified) that will move you closer to being a terrorist. Say, you liked Any Rand’s book Anthem or you are an ACLU lawyer. Well, known terrorists like Anthem, so you move to the middle vertex, where there are two more possibilities. In this case you commented negatively about the US government’s drone program, so you advance to the next vertex in the chain. The last vertex represents the fact you have a Muslim friend on facebook; tah dah! You sir (or ma’am) are a "terrorist." Full story...
Related posts:
(...)
For those of you who are mathematically and statistically inclined, this is what a "decision tree" looks like. If I assign probabilities to each vertex for a possible outcome, I can then calculate the probability of eventually reaching the final right hand vertex. If I don’t "know" the probabilities, I can use Markov Chains to figure out what the right distribution is. If I’ve confused you so far with the math mumbo jumbo, stay with me as I give you an example of the usefulness of this approach. Say I want to know what characteristics "terrorists" share in common, without really knowing. So, I might come up with a bunch of things that I think terrorists share in common with each other: such as "likes" Osama bin Laden, writes about (or is a member of an Islamic group), reads various patriot web sites, etc., etc. I can then seed the model and compare it against the cached web and facebook comments of all of my "known" bad guys and have the model converge on a list of the common descriptors. So, in my mini network in figure 1, vertex 1 (the far left one) is you. The far right vertex represents the "known terrorist" convergence. Progressing frorm you, there are two possible attributes (remember, this is simplified) that will move you closer to being a terrorist. Say, you liked Any Rand’s book Anthem or you are an ACLU lawyer. Well, known terrorists like Anthem, so you move to the middle vertex, where there are two more possibilities. In this case you commented negatively about the US government’s drone program, so you advance to the next vertex in the chain. The last vertex represents the fact you have a Muslim friend on facebook; tah dah! You sir (or ma’am) are a "terrorist." Full story...
Related posts:
No comments:
Post a Comment