"Similar Titles"

[Skavau]

Hi, new here. I have a query on the "Similar Titles" algorithm.

If you click on a title, it says: "Here are the titles which Criticker has calculated to be most similar to [title]. A variety of factors are considered to determine similarity, including genre, cast, year, and shared collections."

1) Does language or country of origin factor in here? It doesn't say it does, but I feel it ought to if not.

2) The "Shared collections" is a good idea here, but it leaves me wondering. Not all collections are equal. People make collections (lists) for shows and films they just happen to like. A ranking of enjoyment. It wouldn't necessarily mean that the entries in that list have anything that bind them together in terms of setting or themes.

For instance, someone could make "Vampire" themed list for the site, where anything about Vampires is added in - and add in True Blood. But some other users could also make a "My Favourite TV shows" and add also add True Blood to that list. To me, the influence of the Vampire themed collection should have much more influence on what is considered similar to True Blood than the list of favourite TV shows by a few users because there's something that objectively binds the entries in that list, rather than a users personal preference. Does the amount of stars a collection gets have any influence here?

Do they have equal weighting? If they do, I do have a proposition to change that if it's wanted.

[geohawk]

Just a note that the "cast" aspect of the similarity model appears to go a long way towards the country/language factor that you are asking for, particularly, since "cast" seems to include directors/writers/creators as well as actors. Especially early in a career, someone is likely to work on projects that are in the same country.

[Skavau]

--

[Skavau]

Just a note that the "cast" aspect of the similarity model appears to go a long way towards the country/language factor that you are asking for, particularly, since "cast" seems to include directors/writers/creators as well as actors. Especially early in a career, someone is likely to work on projects that are in the same country.

Yeah, cast definitely does. I noticed that.

But my question in the main relates to lists. I don't think all lists should be treated as equal when it comes to "influence". Some people make public wishlists, which is fine, but it doesn't mean that the titles in them necessarily relate with each other.

[mpowell]

But my question in the main relates to lists. I don't think all lists should be treated as equal when it comes to "influence". Some people make public wishlists, which is fine, but it doesn't mean that the titles in them necessarily relate with each other.

That's correct, and we don't weight all lists the same. Collaborative lists rank higher, and things like wishlists much lower.

We're looking at other tweaks that might be needed -- "Year end" lists seem to throw things off a little, as it groups disparate films together just because they were like, good, or nominated for awards.

[Skavau]

But my question in the main relates to lists. I don't think all lists should be treated as equal when it comes to "influence". Some people make public wishlists, which is fine, but it doesn't mean that the titles in them necessarily relate with each other.

That's correct, and we don't weight all lists the same. Collaborative lists rank higher, and things like wishlists much lower.

We're looking at other tweaks that might be needed -- "Year end" lists seem to throw things off a little, as it groups disparate films together just because they were like, good, or nominated for awards.

Well I would add that many other collaborative lists aren't built around the idea of binding titles together. Examples I can see now: "Available on Youtube" (two things being on YT doesn't make them similar), "Great Japanese Films" (an opinion list), "Fav Japanese films", "FIlms available in HD", " "Directed by sibling" etc, "#1 at the US Box Office".. even stuff like "Based on a Book" is pretty useless in terms of accurately grouping titles together. There are also a lot of network-based collaborative lists too (as in lists that refer to Netflix, Apple TV, Amazon, CBS etc productions), which I would argue should only have a minor impact on influence.

Do you manually weight these lists?

I'm making, currently, non-collaborative lists for TV series in an attempt to bind them with similar shows based on setting, themes, genre, etc. I chose non-collaborative currently at least because I feel I need a level of control over the content of the lists and there's always a scattershot effect when you open up specific list concepts to the public due to people not understanding the list properly, or simply completely misinterpreting the terms I use.