I want to build my first recommendation system for my thesis.
I see in internet there are many tutorials for making recommending systems for movies, which use these approaches:
User Based Collaborative Filtering
Item Based Collaborative Filtering
but none of these makes sense to me unfortunately if I want to join two users based on their interests,
Is it possible for instance that these approaches instead of recommending a movie, they recommend a user? like based on interests as datapoints and then this tool show you a similar user?
Could you please recommend me tutorials, or an approach for accomplishing this?
I have knowledge of python and machine learning but never did a recommending engine.
At a basic level, I would think that it wouldn’t be significantly different from recommending a movie. The question is just how you get sufficient data about your users.
Well all you really need is some target-values to deal with.
Assuming you don’t have a massive database for user-behavior on dating sites, you can’t utilize ML.
Instead, you will need to define your own metrics on how to match people and just implement them in code.
I think the usual approach is to have a bunch of questions, turn those into a generalized score for a couple of personality-traits according to some basic psychology, turn those into some fitting mathematical structure (like an array), write a function to determine similarity and then just bruteforce that over all user-pairings with a threshhold for when to recommend that.
There surely are some advanced statistical methods as well as other programming-tricks to optimize this. But for a start, going by similarities seems like a good first approach.