erengy / taiga

A lightweight anime tracker for Windows
https://taiga.moe
GNU General Public License v3.0
2.12k stars 229 forks source link

Poor suggestions for titles Taiga can't identify #1033

Open Noctiphobia opened 3 years ago

Noctiphobia commented 3 years ago

Whenever Taiga is unable to identify a title, the list of suggestions is almost always wrong, and also pretty surprising.

An example: Filename [SubsPlease] Shin no Nakama - 01 (720p) [C98D24A8].mkv Suggestions I got: Taiga_gd6wxvRiiv

Using the search, the expected result (Shin no Nakama janai to Yuusha no Party wo Oidasareta node, Henkyou de Slow Life suru Koto ni shimashita) is there (in my list). On the other hand, most of the suggested titles aren't among the results of the search.

I'd expect the list of suggested titles to be basically the result of the search, with titles on my list shown first, or something like that - although I might just not understand the intent behind the current behavior.

paulo27ms commented 3 years ago

What tracker are you using?

Noctiphobia commented 3 years ago

I'm using Anilist, although I remember that this issue was happening at least to some extent years ago when I was using MAL as well.

erengy commented 3 years ago

Taiga uses a combination of different metrics (trigrams, longest common subsequence, Jaro-Winkler distance, Levenshtein distance, etc.) to compare the titles and display suggestions. I don't think it's as bad as "almost always wrong", but I haven't touched the algorithm for many years and there's definitely room for significant improvement.

That being said, this specific example is pretty much an edge case. Since the title is extremely long, the anime gets a relatively low similarity score (~0.53) and is ultimately not suggested. Perhaps the algorithm should give even more weight to overlaps at the beginning, though it wouldn't be wise to change it without testing.

I'll leave this issue open, so let me know if you encounter other cases where you think suggestions could be improved.

nonodesu commented 1 year ago

Another example: image It's understandable that changing the algorithm may be a lot of work, how about you add a quick way to associate currently playing titles with search results? Somewhere in here maybe? image

Noctiphobia commented 1 year ago

I forgot about this issue until I found this - here's a much worse example that I think illustrates the problem the most: Maburaho is the full title of the anime (so clearly not a case of a file having a very small subset of a super long name), and yet it's not in the list of suggestions. This definitely seems like a bug. image

nonodesu commented 1 year ago

Bump. This is very broken. image

paulo27ms commented 1 year ago

Bump. This is very broken. image

That's because the show didn't air yet.

Noctiphobia commented 1 year ago

Here's a case where there are no issues with the show itself, and the name is a perfect match. There's no way this is not a bug.

File name: [Chihiro] Tensei Oujo to Tensai Reijou no Mahou Kakumei 01 [Blu-ray 1080p HEVC FLAC][1BC792E9] The pop-up: image The results in "Not in the list? Search for more." image

Actually, is there any reason for those two lists of results to be different?

paulo27ms commented 1 year ago

Your example detects fine for me. Check that you don't have alternative titles on the series (or have mistakenly added that name to something else that could lead to confusion on detection).

Noctiphobia commented 1 year ago

I double-checked, and there's definitely no such configuration. But also, I only learned about this anime just before downloading it, and I only added it to my list after the pop-up when watching the first episode. One thing to note is that the detection is only broken for me when the title is not on my list. After I add it to my list, it works perfectly fine.

paulo27ms commented 1 year ago

Taiga is using its cached anime list, you should refresh your seasons or use the search as it suggests, that should fix your issues in the future.

nonodesu commented 4 months ago

Taiga_89O2wsdwqz Taiga_TZ6NzqPRGd