F

5 Messages

 • 

110 Points

Wed, Apr 27, 2022 12:19 AM

In Progress

How do you use the Ratings to teach the Recommendations (Top Picks) Algorithm

Seems like I have destroyed any value in the recommendations engine, due to trying to improve it, by rating some of the recommendations. ie, like you would teach the watchworthy app.

I rated some older crime TV shows at a rating that was suitable at the time, eg, I rated a 30 yr old show perhaps a 7. Now, ALL the recommendations are for crime and comedy shows from 20-30 years ago ... pretty useless.

The algorithm seems to value ...

1. Because you watched this, then ....

   BUT, there is no option to say I DIDN'T watch it and HAVE NO INTEREST in EVER watching it.

2. Rating of 7 seems to indicate you like this ...

   WRONG! I liked it AT THE TIME. If I was rating it to indicate my preferences to wishing to watch similar shows NOW, then I would rate it only 1.

Anyone got any ideas how to use the rating system to get meaningful recommendations. Black box stuff is useless - valuable IP, my arse :)  - no feedback or iteration = no improvement.

If nothing else, some sort of steer about the algorithm would be REALLY useful from IMDB, because at present, it's pretty useless.

Champion

 • 

4.4K Messages

 • 

112.2K Points

2 m ago

When a title appears in a recommended list, it has a small "i" in a circle. If you click on that, you can click on a Not Interested icon.

5 Messages

 • 

110 Points

2 m ago

Wonderful. Thanks for the info.

The "Not Interested" functionality doesn't seem to be in the web app, which is what I use by default.

IMDB need to add that functionality there and ...

Promote the information on the website/app, so that more people don't just click 1 to get rid of recommendations.

Other people have been caught out as well, and there was a full explanation provided in this response, but NO information added to the website / FAQ

https://community-imdb.sprinklr.com/conversations/imdb-app-iphone-ipad/mark-titles-as-not-interested/6205a95a6593b177971481b4?commentId=6208c85416676377f25776e5&replyId=62093452116db439b7105d84

Copied in @Col_Needham

Many thanks.

Employee

 • 

4.8K Messages

 • 

149.1K Points

@Flagship99​  Thanks for the feedback — this functionality is now available on the web site.  The same “(i)” appears in the bottom right corner of each title card in “Top picks” on the home page (due to a bug, it does not appear in the equivalent section of the What to Watch page which we presume is where you were looking;  we have opened a ticket with the appropriate team). 

Here’s how this section of https://www.imdb.com/ looks for me with the first “(i)” clicked in “Top picks”

Hope this helps. 

5 Messages

 • 

110 Points

@Col_Needham​ Brilliant! You're on it.

Any chance of an FAQ about the issues and considerations of recommendations and getting the best out of the IMDB recommendations algorithm (and ensuring that dopes like me don't screw up the ratings system in the process).

Many thanks for the help.

5 Messages

 • 

110 Points

@Col_Needham​ I still have some serious misgivings about the IMDB recommendations algorithm and it's antagonism to the ratings system.

For example, I have watched hundreds of movies which I haven't already rated in IMDB. If they show up in Recommendations and I really enjoyed them, then it would seem reasonable to rate them.

This entirely ruins the Recommendation algorithm which then gives me 10 more films like that 30 yr old romcom as recommendations.

It would seem that Ratings and Recommendations are NOT compatible.

It would seem that you cannot rate historically watched films if you want the Recommendations to work for you.

It would seem that in Recommendations-Hot Picks (on the homepage until it's offered in What to Watch page), that you need to annotate as Not Interested all films or tvshows that you may have watched previously but which you do not wish to have a similarly aged (30yrs ago!) recommendation made.

Would this be a good summation of the situation?

Many thanks.

uomotorta

31 Messages

 • 

542 Points

2 m ago

I would like to understand better how the algorithm works.

I do not make heavy use of the recommendations, because I find it quite useless.

It seems to me that I can just find movies with actors and directors of my most recent ratings.

It doesn't really matter if you liked your most recent movies or not.

It's basically what you find in the "more like this" section in the title page of movies.

For instance, if you watch 3 movies of Rocky in a row and you don't like them at all (eg rating 4), you will still have other Rocky or Rambo movies.

But maybe I'm wrong.

(edited)