Suggestions

[aegisx]

I was a fan of Tivo's suggestion system, since I often didn't know what I wanted to tape.

Anyone have plans for something similar?

[joche]

What is that? Please assume not everyone to be a US citizien!

[reven]

i think it suggestion which shows to watch/record based on your previous recorded/watched shows.

[jasonf]

INHT (I never had Tivo), but I know that there's a "thumbs up" and "thumbs down" button so that you can rate programs that you recorded. Then it does an Amazon.com-like thing, and suggests other programs that you might like based on your previous activity and the activity of others with similar ratings for the shows. I think it even goes so far as to automatically record some of those during idle time.

I briefly looked into this yesterday, strangely enough. One source of data might be IMDB. If you look up a TV show in the database, there's a Recommendations section. For example, here's the recommendations for "Lost" (ABC Wednesday Night show that debuted this past fall):

http://www.imdb.com/title/tt0411008/recommendations

So, lets say that I gave "Lost" a Thumbs Up rating. I then gave another show another thumbs up rating. If both of these shows had "Alias" as a recommendation, then "Alias" would be higher on the suggestion list (unless, of course, I give "Alias" a thumbs down).

Something like this would work on the Series level of data, but probably not Episode-specific because of the scope of data in IMDB. Meaning, let's say that I liked one episode of "Lost" because of a certain actor or topic. I believe (again, no personal experience to back this up) that Tivo will take that into account and try to match metadata about the episode to generate better suggestions.

Example: If I keep giving high ratings to all episodes of "Mad About You" that Hank Azaria appears in, then "The Simpsons" might be a logical suggestion because Hank does a lot of the voices.

[aegisx]

Thats pretty much how it works. So after you give a bunch of shows ratings (or if we don't want to rate shows, it could just make recommendations based on what you've previously recorded). So if I tend to record sci-fi shows, and it sees a sci-fi show on, and nothing is recorded.... it will record it. I found a bunch of shows I would't have though to watch based off of that system.

[jerichar99]

That IMDB page jason mentioned would probably be the easiest way. I use IMDbPY as the command line interface for IMDB; I'm not sure if "Recommendations" is part of IMDbPY yet (I've never needed to look at Recommendations).

To do episode-level suggestions via some stand-alone algorithm would be challenging. Sure, there are some easy matches like the actors in the episode. But beyond that, it probably wouldn't work very well.

The system that could work (but could be considered very invasive) is to pool together users' liked and disliked episodes. Then when you start marking likes and dislikes, it can compare those to other users'. Thus, if user A likes episodes X and Y and dislikes Z, and user B likes X and dislikes Z, then there's a possibility that user B will like Y.  Of course, in a real system, you would be comparing more than three episodes, and perfect matches would be a rarity (so perhaps some scoring system would be used instead of looking for perfect matches).

So perhaps the system can be a buy-in style, where if you provide your likes/dislikes, you get access to the other users' data.

[joche]

Thanks for the explanations!

I guess the only really working solution would be to have a MASSIVE userbase and create suggestions based on that. I mean, not even Autonomy ($$$ seach engine) can create good enough suggestions just based on text masses.

It's better to go the Amazon way and create suggestions based on other users ratings. I once built such a solution to rate documents on an intranet and it worked out fairly well.
It worked a bit like this:

User A rates a document five stars (1-5 is possible)
User B rates the same document five stars

User A would the be be presented with other documents that user B had rated with five stars, assuming that they have the same interest.

(Well, the algoritm was a bit more complex than that, but anyway - you get the idea).

So, to get this up and running I believe someone must host a service where user data can be posted and processed, that would be the toghest challenge for a community based project as this.

[jasonf]

I would say that this problem/feature set is not unique to GB-PVR, so there may be interest from the other PVR groups out there to pool together on something like this.

It would probably conflict with existing patents, etc, but I would go so far as to say that this would even make a great startup business (to provide this service for the PVRs). I'm not in favor of charging users for usage--Tivo and ReplayTV already do that. However, the aggregate anonymous data could be sold to whoever wants it in order to support the infrastructure overhead (hosting, bandwidth, support staff).

Just an idea.

[KingArgyle]

You guys might want to check out RatingZone.com, they already have a category setup for TV shows, and allows you to rate those shows. They will then give you a list of recommendations based on the shows you rated.

http://www.ratingzone.com

Somebody might be able to provide a screen scraping interface to this site, or even the IMDB site to pull the information. It'll probably be quicker to do that, instead of trying to build up your own custom interface. If the interface was designed as a Web Service, then other PVRs could also use it. It would then be on each PVRs development community to interface into the system.

[aegisx]

So is genre not part of the meta data at this point? I wonder if the full set of meta data would contain all the necessary info. I know sub isn't including it all because everyone cannot use it.