Jump to content

Ever thought about (ML.Net) Machine Learning Media Recommendations?


chef

Recommended Posts

chef

Hey chef,

 

Will the plugin allow an opt-in for every Emby user or will it simply collect data of all users?

 

 

Is there a way to like a movie with Emby for Kodi? @@Angelblue05

Sure, we could have the admin choose users to opt in.

 

Because it would be a server plugin, it would gather favorite from all clients, yes.

  • Like 1
Link to comment
Share on other sites

horstepipe

Perfect,

getting this to work would be a real game changer.

I‘m really excited about it!

 

Thanks for your effort!

Link to comment
Share on other sites

chef

I've spent the last week learning quiet a bit of ML.Net and what it offers. It's quite powerful. But, the Classifiers/Models are only as powerful as the dataset they consume.

 

That said, as cool as it would be to teach an algorithm to recommend media through emby, the task itself is massive.

 

The dataset would have to contain a world of media items, with just as much user data.

 

Unfortunately, it is probably best left up to companies like Netflix and Amazon who have all that data.

 

For example, I spent the last week collecting Home Automation data for my house.

 

I saved it in lists of JavaScript object notation. (What devices where powered up, etc.) I logged device data ever minute for weeks. Then serialized all the data back into a Binary Classifier, I trained the thing and still only got a production accuracy of 0.5.

 

So, for that project I'll have to go through the data and figure out what is useful and what isn't to get a better rating.

 

I only have 24 devices on the mesh network.

 

Now, I imagine a dataset of media items (TV, and movies) and realize the massive undertaking.

 

The Machine Learning Model is only as good as the data it consumes...

 

Perhaps in the future, machines will have a better way of sorting and consuming datasets. They'll be smarter and figure it out for you. In the meantime it's true: Artificial Intelligence/ Machine Learning (because there is a difference between the science and the application) is just not that intelligent.

Edited by chef
  • Like 1
Link to comment
Share on other sites

chef

Well, I did it, I created my first Machine Learning application.

 

I recorded over half a million home automation devices states over the past two weeks, fed them to a BinaryClassification Learner, and gave it control of my house!

 

Since yesterday it has done pretty well. It knows weekday routines really well, it is just the weekend routines it needs to predict better.

 

It's really just an OP light timer, but... it taught itself how to be that way. LOL! 

 

Pretty crazy!

Edited by chef
  • Like 4
Link to comment
Share on other sites

BillOatman

I still like your original idea, and I think ultimately it would have done a good job with a seamless plugin to collect data to a central database for a learner to use :)

  • Like 1
Link to comment
Share on other sites

  • 3 weeks later...
mrfragger

That said, as cool as it would be to teach an algorithm to recommend media through emby, the task itself is massive.

 

The dataset would have to contain a world of media items, with just as much user data.

 

Unfortunately, it is probably best left up to companies like Netflix and Amazon who have all that data.

 

So, for that project I'll have to go through the data and figure out what is useful and what isn't to get a better rating.

 

I think this excerpt from this TED talk with CEO of Netflix Reed Hastings interviewed by Chris Anderson (TED founder) is relevant. Sorry for the length. But in a nutshell people rate in an idealistic way but in reality want to watch not so much what one would tend to typically think. I watched a documentary on all the Oscar winners and the reoccurring themes used ad nauseam but it’s what the award show wants people believe how pure our societies function. So all these algorithms are in tune with what people actually really want to watch but rather try to veer people to watch what Big brother wants them to watch and that’s my issue with this whole shebang.

 

Here’s the TED talk excerpt:

 

CA: You've got this other secret weapon at Netflix, it seems, which is this vast trove of data, a word we've heard a certain amount about this week. You've often taken really surprising stances towards building smart algorithms at Netflix. Back in the day, you opened up your algorithm to the world and said, "Hey, can anyone do better than this recommendation we've got? If so, we'll pay you a million dollars." You paid someone a million dollars, because it was like 10 percent better than yours.

 

 

08:51

RH: That's right.

 

 

08:52

CA: Was that a good decision? Would you do that again?

 

 

08:55

RH: Yeah, it was super exciting at the time; this was about 2007. But you know, we haven't done it again. So clearly, it's a very specialized tool. And so think of that as a lucky break of good timing, rather than a general framework. So what we've done is invest a lot on the algorithms, so that we feature the right content to the right people and try to make it fun and easy to explore.

 

 

09:20

CA: And you made this, what seems like a really interesting shift, a few years ago. You used to ask people, "Here are 10 movies. What do you think? Which ones of these are your best movies?" And then tried to match those movies with recommendations for what was coming. And then you changed away from that. Talk about that.

 

 

09:41

RH: Sure. Everyone would rate "Schindler's List" five stars, and then they'd rate Adam Sandler, "The Do-Over" three stars. But, in fact, when you looked at what they watched, it was almost always Adam Sandler. And so what happens is, when we rate and we're metacognitive about quality, that's sort of our aspirational self. And it works out much better to please people to look at the actual choices that they make, their revealed preferences by how much they enjoy simple pleasures.

 

 

10:12

CA: OK, I want to talk for a couple of minutes about this, because this strikes me as a huge deal, not just for Netflix, for the internet as a whole. The difference between aspirational values and revealed values. You, brilliantly, didn't pay too much attention to what people said, you watched what they did, and then found the stuff that, "Oh my God, I never knew I would like a show about making horrible recipes, called 'Nailed It!'"

Edited by mrfragger
  • Like 1
Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...