Hey guys, so I've been doing some research into belief systems, value dimensions, social choice, and other stuff. This research has led to a pretty nifty way on how to allow users to define their own belief system and value anything. And by belief system I mean your wise wordly views. Anyway, when I started this research I had this idea in mind: "What if there was a wiki where I could look up anything and determine if I think it's valuable or not, and more importantly why" (no more trouble trying to decide which topilet paper to choose at the supermarket). So we (by we I mean my supervisors and I) have been working on this framework and a few algorithms that may be able to get this done. It seems to be going well so far, your input and ideas could help a lot though, so I'm presenting the idea here and looking for some feedback.

It's all based on value dimensions and concepts. Concepts can be groupings of things, 'Universities', 'Cars', 'Restaurants', etc. And each concept has a number of value dimensions attached to it, which are all ordinal values. So for a restaurant you can have value dimensions of 'atmosphere' which could range from amazing to bad, 'menu variety' which can range from limited to a lot, etc. Then under each concept you'd have your various entities, under restaurants you'd have Mc Donalds, Johnny Rockets, My Local Hotdog Stand, etc.

Then you define your belief system by saying you think the ideal value for 'atmosphere' is amazing, your ideal value for 'fat content' could be 'a little fat', etc.

So a user would be able to go into this wiki, add concepts, value dimensions and entities, and then rate an entity over the value dimensions that have been attached to the concept the entity belongs to. It's like all the rating sites you have around these days (ImDB for movies, glassdoor for jobs, etc), but where users themselves can define the value dimensions and which value is ideal. And by allowing users to define a belief system, the framework can know what is globally ideal and with the addition of know each person's personal beliefs, recommendation and rating accuracy should go through the roof, in theory...

Anyway, the test site that has been implemented seems to be working really well. It's just that I do not know how willing people would be to define their beliefs, would people participate enough to make a site like this credible? How would you deal with things like devaluation (price changes over time)? Would you be willing to define your belief system or would that be a major invasion of privacy?

Also another thing I've noticed on the test site, is that people don't really understand how to define their ideal values. I've created a survey for this actually, it's just 6 multiple choice questions, takes 5 minutes, if you have some time I would really appreciate some input there as well.

What do you guys think?

Like NetFlix ratings but for life?

"Because you had interest in Jesus, you might like Buddha"

Something like that, only you decide what dimensions matter instead, using a wiki-like system.

Because i like being the devil's advocate, i'm going to be the devil's advocate:

- Unless you have a very specific niche, you are going to have to define new attributes for every topic you want to categorize. FatContent works for a burger joint, but not for silverware, video games, or a new couch.

- People may value something for attributes you have not defined. Perhaps you need user-definable attributes? Of course when you allow something like that, you just opened Pandora's box.

- Some things are not quantifiable. Art and music are classic examples. I like a song, you don't. You like a painting, i don't. Additionally, the attributes that apply to a specific work of art do not necessarily apply to the entire category of "art" and may be entirely localized to that one work. You may not even be able to describe why you like something, just that you do.

- Tastes change.

- Things change. ("Under new management!")

- You might consider leveraging one of the many standardized personality tests and use that to map to attributes in a way that you define. Then users won't have to constantly be defining their views on attributes, just what sort of person they are.

The devil's advocate can be a good thing ;)

Yes defining new attributes is part of the whole deal, as is user-definable attributes (aka value dimensions). Having to define new categoroies for articles you create in Wikipedia is not a problem. It's the same with this, if you add a new concept, you may need to add new value dimensions.

I would rephrase that to somethings are not objectively quatifiable. They are subjectively quantifiable though. I could give a picture 7 out of 10 on my own personal scale, and you can give it 3. It's just a different type of quantification. And, theory based on wisdom of the corwds says that under certain conditions (all of which this system allows for), you can reach a "theoretical" best. With the addition of allowing users to define their own belief system, you could probably reach a theoretical personal best then.

Tastes change, and you cna change your ratings as well. Things change is a big problem, devaluation specifically... since this whole system is used to assign personal and universal value to things, devaluation is a prime concern, which I do not know how to go about... For example a laptop, when it comes out it would be rated 'expensive'... but this rating becomes void after a few years when it is cheap... maybe some math-play with its release-date can solve the devaluation issue.

Personaliy tests is a good idea. I'm definitely going to look into it. Not right now though. Only problem I can see with personality tests is getting users to actually take the tests...

Along with devaluation, you might also factor in social devaluation; like when fads die out. Remember when Hootie & The Blowfish was OMGGREATESTROCKBANDEVAR!!! ? Where are they now? *shrug*

Algorithmically, you basically score something higher if the votes are more recent, i suppose.

Quote: Original post by leiavoiaRemember when Hootie & The Blowfish was OMGGREATESTROCKBANDEVAR!!! ? Where are they now? *shrug*

Hootie is one of the most successful Country musicians currently...

I think its an interesting idea, might I ask what algorithms are powering the recommendations? however, i think that asking people to fill and maintain these wiki like systems is simply too much. I dont think you'll be able to gain enough traction.

Also you may find that sometimes too much information is not good. In creating a recommendation, having so many parameters may introduce noise in your model when making a prediction in a specific class that it becomes little better than a random guess. If you remember the Netflix competition they were only able to scrape an improvement over the baseline because the problem was relatively simple parameter wise - relatively, it was still extremely complex in and of itself.

Similars: Hunch, DirectedEdge , Amazon + Facebook recent partnership.

Quote: Original post by way2lazy2care

Quote: Original post by leiavoiaRemember when Hootie & The Blowfish was OMGGREATESTROCKBANDEVAR!!! ? Where are they now? *shrug*

Hootie is one of the most successful Country musicians currently...

Does he still get mad when people call him Hootie?

Anyway, how often do people rate something to the highest or the lowest extreme? For instance (and in keeping with the Netflix example), if I can rate a movie up to 5 stars, but my desire, regardless of whatever was technically good about the film, is to never see a movie like that again, then I'll probably rate it 0 stars (if I rate it at all). This extremism creates inflation/deflation, just to satisfy my opinionated desire to not see that thing. And if desire is fleeting, then the ratings are fleeting, and the system is broken.

Quote: Original post by Benjamin Heath
How often do people rate something to the highest or the lowest extreme?

Good point. I think youtube recently went from a 5-star system to a simple thumbs up/down system because people were not using numbers 2, 3, or 4.

And then of course you have idiots who think they are rating the service and not the product/object which pollutes with results.