Continuing with my recent theme of musings. I’ve been musing over this idea of using genetic algorithms to discover content. Imagine a scenario in which you would like to know things that might be interesting that you wouldn’t know how to search for on your own. Lets say science things. Naively this bot would start with a random set of words from the science dictionary and search twitter , and return back tweets. The tweets would have a set of features be that length, avg. word length, time, number of sciency things. our objective function could then be to monitor the number of likes , that each of the tweets get. So with each generation, you might discover or find things that pair certain features together that would weight some more heavily than others , this would the vary the filtered results. And eventually I’ll be it after sometime you’d get a feed that might be interesting to you . The initial randomness and mutations might make the evolution of the algorithm given a good feature space recommend things to you in an unexpected but interesting way.
One step further might be to simply generate a population of virtual users, that tumble content, with the objective function of getting users that don’t normally engage to engage. The features that resonate with some group or segment of users would then cause the attributes of those posts to be weighted higher, and the virtual user to generate or pull rather better content.
Genetic Algorithms are most definitely not a panacea, but i think I’d like to try harder think about them in terms of searching a space in an interesting way or generating content.