I have an idea and I'm trying to figure out a formula that would best achieve the goal.
I have a large set of data and each data point has a few dozen parameters. Some parameters are boolean, others have a small range (E.g. 1 to 10) and others have big ranges but I can compress them all down to a float between 0 and 1.
I'd like to show a few data points to a user, and get their feedback on which data points most closely reflect their preferences. The simplest would be a up vote or a down vote for each data point.
I know a subset of parameters will have a significant influence over others for all users, but the less important factors may be significant for an individual user, so I can preset a few of the weights accordingly.
It needs to both identify which parameters are most important, and the ideal value for the parameter. That way I can calculate a vector to other items in the dataset and present the closest ones.
I don't think this is getting in to full blown neural network territory, and I'd like to find a solution which can figure out likely parameters with a very small amount of guidance from a user (e.g. 5 up or down votes). Accuracy isn't critical, it's more to point the user in the right direction. I'm just not sure exactly the right approach.