"Letting fatties roam the site is a direct threat to our business model"
news.bbc.co.uk
news.bbc.co.uk
Sorry that was more of a rant on the subject of self promotion through shady means than a commentary on the actual article.
http://img39.yfrog.com/img39/6247/h807.png
On a more serious note, there are a lot of difficult problems in onlne dating that someone could make a lot of money by solving. The lack of attractive pictures is not one of them.
1. Limited corpus. This is usually the result of sites going after very specific verticals (i.e. race, religious, <some-other-facet> specific).
2. Terrible search. Granted, determining compatibility isn't easy, but you typically can at best search on dimensions to limit gender, sexual orientation, proximity and age. After that... then what? If you're using a site with a large corpus (Match, PoF, etc.) you end up subconsciously surrendering your standards and resort to "clicking on the hot ones".
I would liken search on dating sites to be similar to web searches before Google. Actually, before AltaVista. It really is that bad.
If compatibility could be determined quickly and reasonably accurately, on a large number of dimensions, one company could crush every other dating site like an ant. Especially if that score is used all over... for instance, your mail could be sorted by potential compatibility - so you don't waste your time looking at messages from individuals that aren't remotely relevant.
The problem is that sort of quantitative analysis is very computationally expensive in a high-dimensional space, especially for a large corpus (say, 5 to 10 million users).
>which is why surveys and "shared values" questionnaires fall flat Bingo.
Surveys actually provide a way of gaming the scoring system as well, since you can make a reasonable assumption of what your ideal partner would want you to say.
That's the biggest flaw with dating sites, is that you can only trust so much user input - everything needs to be taken in context. Even simple things such as "interests" should be evaluated in their own context, because you can't rely on exact matches or even fuzzy matches - you have to infer what else their might be interested in. Granted, you also get the problem where you don't get enough user input, therefore being able to infer other information becomes ridiculously valuable.
I agree that you can't perform N^2 searches in real-time. A lazy match is fine until a more accurate index is generated. However we still lack a reasonably accurate index. We also have to do all the searching ourselves, instead of the system working harder to provide far greater relevancy than it does now. I think the best analogue is Netflix - I want recommendations based on my interests and behavior. Again, the issue is dating has far more dimensions than movie recommendations.
I would agree that most dating sites suck, but surprisingly it turns out one of the free ones is pretty decent. (Being free helps deal with the limited corpus issue nicely as well.)
I think they solves the high-dimensional space issues by lazily evaluating match scores and caching. I have no idea what they're doing on the backend, but I'd guess they don't actually calculate scores for all n^2 matches everytime someone answers new questions.
The computer science behind the site is frankly fascinating. While the site itself doesn't take itself seriously, they seem to take the algorithms driving it very seriously.
The bottom line is it's worked for me. I'm not someone who can't talk to people in person, I'm just someone who's busy and found it more efficient to meet people online. For me what exists now is actually workable, not limiting and efficient.
I'm sure it could be improved, but the two problems you cite haven't been that problematic in my experience.