Switch to full style
Discussions about Coollector.
Post a reply

Reliability of the predictions

Tue Aug 07, 2018 2:57 am

Let's be honest, we never know for sure how much you'll like a movie. Sometimes we have a lot of clues pointing towards a predicted rating, and sometimes we have less clues. In other words, some predictions are more reliable than others. A prediction with a low reliability isn't necessarily wrong, but a prediction with a higher reliability is more trustworthy. The new version 4.11.4 makes it easier to view the reliability of the preditions, and an advanced setting allows you to adjust the minimum reliability below which a prediction is kept hidden.

You'll find many more details about our recommendation system in the (updated) help section:

https://www.coollector.com/help.html#recommendations

I hope you'll like those improvements. They make our recommendation system even more unique. To my knowledge, nothing comes close to it!

Re: Reliability of the predictions

Wed Nov 27, 2019 1:33 am

I'm somewhat new to using Coollector, but am very much enjoying it! I've rated over 400 or 500 movies, and was wondering about the reliability number. Where do the reliability numbers come from? I understand that the higher the reliability, the more trustworthy it will be, but wasn't sure where the numbers come from. Thanks!

Re: Reliability of the predictions

Wed Nov 27, 2019 2:36 am

djzimm wrote:I'm somewhat new to using Coollector, but am very much enjoying it! I've rated over 400 or 500 movies

Cool!!!


djzimm wrote:Where do the reliability numbers come from?

Have you clicked the link above? I've explained it the best I could, please tell me if something isn't clear enough.

https://www.coollector.com/help.html#recommendations

When the collaborative filtering algorithm finds a lot of correlations to analyze, it can produce a prediction with a high reliability. But the reliability may be low because you haven't yet rated enough movies for the system to really understand your tastes. Or because the movie isn't popular enough among the other users to get enough hints about how much you'll like it. Or maybe you have very special tastes and it's hard to find other users with the same mindset (as an extreme example, if you rate completely at random, the prediction engine will of course have a hard time to understand you).

Not all predictions are equal, obviously, and I'm proud that my system is open about that. I don't know any other system which discloses that kind of information. I'm truly passionate with recommendation systems, and I hope I didn't go overboard with that feature. It may sound a little bit technical and I hope it won't scare away people (discard it if that's your case), but in my opinion it's a fundamental aspect of the predictions, and very useful too! For example setting a higher reliability treshold (in the settings) can clean up your list of recommendations by keeping only the most essential recommendations. The realiability is also used by the Rating Helper which will ask you to rate first the movie with the highest reliability, because it's the movie that you're the most likely to have seen.

In all modesty (and until I'm proven wrong), my system is the Rolls-Royce of the recommendation engines, which isn't a small feat because there are a lot of big players in the field 8)

Re: Reliability of the predictions

Sun Dec 01, 2019 12:32 am

Thanks for the reply! And yes, yours truly seems to be the Rolls-Royce of rec. engines. :D Continuing with the reliability feature, of course everyone will have their subjective opinion. But if I've rated around 600 movies with some shows too here and there, have you heard what others like to bump the minimum reliability up to? I've played around with different minimums, but just curious. Thanks. ;)

Re: Reliability of the predictions

Sun Dec 01, 2019 1:14 am

djzimm wrote:And yes, yours truly seems to be the Rolls-Royce of rec. engines. :D

Thanks, man!


djzimm wrote:have you heard what others like to bump the minimum reliability up to?

Unfortunately I haven't, I can only speak about my own experience...

I've rated a little above 2,500 movies and shows. When I want to know what are the most recommended movies for me, I set the minimum reliability to 1,000 and this gives me about 4,000 predictions (you'll have to settle for a smaller number because you've rated less movies). But I only do that from time to time. The rest of the time, when I'm deciding what to watch on TV or streaming, I set the minimum reliability to only 100 so I can get a prediction for nearly everything that I have access to. If the reliability is low, I also take into consideration the IMDb rating. When the movie is brand new and the IMDb data is all I have, sometimes I go for it because the movie has a big appeal, and sometimes I wait some weeks to see what the recommendation system will tell me.

Re: Reliability of the predictions

Sun Dec 01, 2019 9:34 am

Big help for equating that to where I'm at! Thanks! ;)

Re: Reliability of the predictions

Mon Dec 02, 2019 10:30 pm

Apologies for my question unrelated to reliability! But is there any post here for resolving publishing online issues. I tried twice and after waiting a few minutes for it to publish, the web browser was white with an error message.

Re: Reliability of the predictions

Tue Dec 03, 2019 10:50 am

What was the error message? I've just tried it and it worked fine for me, even with my 2,500 rated movies and 40,000 predictions.

Re: Reliability of the predictions

Thu Dec 05, 2019 5:32 am

I could be doing something wrong, but the phrase is, "ERROR: invalid parameters."

Re: Reliability of the predictions

Thu Dec 05, 2019 6:41 am

Did you set a privacy key? Maybe it contains a character that the script can't handle. If you find out it's the case, please tell me which character it was.

Re: Reliability of the predictions

Thu Dec 05, 2019 9:20 am

So I just switched the published data from predicted and rated persons to predicted and wish list and it worked online. :D Also, if you're still wondering, the key was never used or was a problem. ;)

Re: Reliability of the predictions

Thu Dec 05, 2019 11:05 am

djzimm wrote:So I just switched the published data from predicted and rated persons to predicted and wish list and it worked online. :D

Sorry, I didn't understand everything... Which checkboxes you've unchecked?
Post a reply