4 ms·
I genuinely believe that you're making a mistake over here; you're assuming that machine learning is a magic bullet that can solve anything. I am afraid that is
by todayiamme 16y ago
I genuinely believe that you're making a mistake over here; you're assuming that machine learning is a magic bullet that can solve anything. I am afraid that isn't the case. Even for a satisfactory level of functioning you need to factor in these "meta" variables;
- Emotional stability of the user: You need to be able to judge if the user can form serious long term relationships before recommending him or her.
- Level of Emotional maturity: Maturity is really different from stability. Stability means that you'll be roughly the same a week from today whereas maturity applies to your ability to handle problems effectively and rise above petty issues.
- Willingness for Engagement/Commitment: Your users are in a portal, but the levels till which they are willing to use your service vary, and it isn't something you can ask in a questionnaire.
- Acceptance: This is a part of emotional maturity, but it deserves a solitary mention. How do you judge if someone is willing to accept divergent points of view and learn from things?
- Perspective: What does the user seek qualitatively in relationships?
I think that one day this might be possible, but you're dealing with one of the greatest unsolved problems we have ever faced. Making a computer understand things.
Further, I believe that even though you might succeed someday it won't work for people all the time. Why? Because emotions are really hard to parse, and what I seek in relationships is far beyond common interests or anything like that. I seek kinship, love and understanding. How can you possibly judge and match that without having some degree of emotional understanding? This is yet again one of those great unsolved problems.
On the other hand, there must be a solution beyond simply matching interests from a long checklist. Maybe you should approach it like Aardvark. You could make a friend of a friend based contextual system. Imagine a giant grid of people remotely connected to each other through friends and you parse the daily routine conversations between users and create a personality profile using current psychometric testing (it's seriously inaccurate, but we need a shot in the dark). After that you offer to match them together anonymously at first (this is important) and then see how they hit off.
Yet again I doubt that it will work at all due to privacy issues and moreover it will be too computationally expensive to justify...
- pinksoda 16y agoWhy would you need to go through all of that for a FRIENDSHIP? I don't think you read his post.
- todayiamme 16y agoTrust me I read his post. Any relationship I engage in I take it very seriously. I want to be there for people and love them. That's what it means for me. On the other hand, I was trying to list what happens in someones mind when they try to decide if someone should be close to them or not. The list must be pretty inaccurate, but it was more to show a point than to form a blueprint. P.S. - I know that I tend to take stuff too seriously, but I just need to emotionally connect with people. So, I guess that YMMV.
- scott_s 16y agoI think most people do, just unconsciously. We don't keep a mental check-list going in our head. But if after interacting with a person for a while, we can say "I dunno, something about him just rubbed me the wrong way," then they failed an item on our implicit list we may not be conscious of.
- cb33 16y agoYeah, I think this is a little too complicated. Only people who are looking for friends would be on the site and I think if you supplied them with a list of people who have similar interests, are around the same age and describe themselves similarly, then people should be able to find friends. Of course, you would need to have a huge userbase.