Metadata download matches too generously

[sarasas]

I just upgraded from Calibre 2.75-ish to 2.81 and ran a batch metadata download on items in my database with no previous match and it went horribly wrong.

A lot of entries got matched of with completely unrelated books based on the title name being mentioned in a comment to another book somewhere. Those entries got updated with new titles, authors, tags and comments and were basically replaced with a different book.

I went through them manually to try to restore them and noticed that for some of the book, the right data was available, just not as the first match in the list.

I think matching titles against comments and descriptions of books is a mistake but I cannot find a way to turn it off. I hope this is just a bug that can be easily fixed.

[kovidgoyal]

There has been no change to how metadata is matched, the only change is that now instead of searching amazon directly, amazon is searched via a search engine instead. Presumably for you problem books, the search engine's top link is not to the book page but to another book.

You can get back the old behavior by customizing the amazon metadata download plugin to download from amazon's servers, but be warned that doing so might result in failures because amazon is trying to block calibre.

[kovidgoyal]

And post the title and author for some of these problem books, and I wil see if I can tighten the results a bit.

[sarasas]

An example is my entry named with
title: "Bitter Almonds"
author: "Dorothy L. Sayers"

Download metadata searches "Goodreads, Barnes & Noble, Google, Amazon.com".
It gets 2 hits, both from Amazon.com. One is "Bitter Almonds: Recollections and Recipes From a Sicilian Girlhood" by "Maria Grammatico" and the other is "Bitter Almonds : The True Story of Mothers, Daughters, and the Seattle Cyanide Murders" by Gregg Olsen. Neither is right, but is at least in the same general area.

A more extreme example is an entry
title: "Nebuchadnezzar"
author: "Dorothy L. Sayers"

Here the download matches it to "Psalm 119: The Diary of a Captive" by "Gene Cunningham". The description of that book contains the text "This study of Psalm 119 is based on the assumption that the author was one of the captives taken in Nebuchadnezzar's third and last deportation of the Jews to Babylon" which includes the string Nebuchadnezzar.

[Eve1972]

I have been having the same issue actually. I always batch download metadata for the books I import and before I updated it would accurately fetch the correct book info 98% of the time. Since the update, that accuracy has dropped to 50% if that.

I am going to try the server trick and see if that helps!

[sarasas]

kovidgoyal, maybe Calibre could do some internal matching of title and author after fetching the results from the search, and if they don't match, discard them? Because it does seem the search interface is a bit too "helpful".

[kovidgoyal]

Yes, that's what I meant when I said


"And post the title and author for some of these problem books, and I wil see if I can tighten the results a bit."

[kovidgoyal]

This should take care of it: https://github.com/kovidgoyal/calibr...b56f98687efc7b