Author Division: Philosophy

This topic picks up on the announcement of author division.

Some first thoughts. Let's discuss.

Basic idea. LibraryThing is ultimately based on members books. Although the "global level" can be edited to some extent, it's existence generally "bubbles up" from members' books.

Member books come with authors, but they are by and large just names (ie., "strings") to us. Names alone aren't enough to differentiate between the various Richard Wagners and Steve Martins out there.

Libraries solve the problem by, basically, changing the names. They add birth and death dates, and other factors, like where authors were born. The library method stumbles on the artificiality of these names, the extra work required to get it and keep it updated, on the fact that only library data has it, the lack of consistency in library data, the lack of a single authoritative "authority file," and other factors.

LibraryThing's method(1) is to split authors by something we already know--the works. This allows members to divide without inquiring into author dates(2)

A legacy complication. Unfortunately, LibraryThing is captive to legacy data and structures. For authors the problem is that the global author layer isn't actually composed of names and works, but of simplified forms of the name. These simplified forms, which look like "twainmark" and "careylisa," have various limitations, like being a maximum of 20 characters and not being able to distinguish between, say, Robert A Hall and Roberta Hall.

We added rudimentary author separation some time ago, but it didn't involve secondary pages. My plan was to fix this first then move to separate author pages. Eventually, however, I decided that members' difficulty to add secondary data, like Common Knowledge, outweighed the extra work that would now be necessary, and the limitations of the current system being extended still further.

Names vs. Splits. We're going to need to think through all the issues with combination and separation--which involves names--and division--which involves works.

For example, right now, members combine names (actually simplified names) to make authors. So, Mark Twain is composed of twainmark and secondary names like clemenssamuel. These secondary names get pages of sorts.

Division adds complication. For example, John Wilson is divided into fifty-nine authors, and includes 10 names. Some of these names, like "Rev. John Wilson," only apply to some of the divided authors—not all of the John Wilsons were in fact pastors! In theory, we could start encouraging people to combine names into divided authors—to combine "Rev. John Wilson" into John Wilson (45) alone, or perhaps into John Wilson 1, 45, 23, 24 and 45.

My tentative solution is this: Name combination happens at the top level alone, and only the top level shows these alternative names. Once these names are combined together, they imply a list of all works associated with the names. Then works that result are split out from them. "Also known as" are not shown for divided authors; the complexity and uncertainty are too great for that.

---

1. As with all the LIS-type ideas LibraryThing has pioneered, expect it to appear in all our competitors, without credit.
2. This is a good thing. LibraryThing does allow members to add dates, and we frequently get authors calling us up demanding we remove the dates. Very often we got the data from the Library of Congress authority file, and we plead in vain that the same information can be found in thousands of library catalogs around the world...

The problem of 1, 2, 3...

We've never really defined how the 1, 2, 3, etc. should be divvied up. I believe I've said that #1 should be the most popular author, but we've never made it a hard rules.

I think we need to do something like that. There's something weird about John Wilson having #1 with only 5 copies of his books, and #2 having 4. Meanwhile 4, 8, and 12 all have over 100 copies. I'm guessing the intent was to order them by time, but I don't absolutely know.

The numbers are semi-static. It doesn't recalculate who's one every time and I don't think it ever will. That way lies chaos, I think. So, what shall we do?

Do you think it would help to have a space for a freeform (CK-ish) string to display as well as or instead of the ordinal number, and act as mnemonic for the sub-authors? "John Wilson (Mexico, b.1942)", or "John Wilson (cartographer)". Then the 1,2,3 order doesn't matter as much, and managing the splits might get a bit easier.

2:
I think now it should be able to recalculate the numbers. If we get the ability to combine John Smith (5) with John X. Smith (fake example) then there's no reason for the other John Smiths not to shuffle up to fill the gap.

Wait, no, that's not exactly true... How will new works attributed to John Smith that should be on John X. Smith's page get split off to the appropriate place?

Recalculation is doable, but it will have to take place offline or when nobody can be reasonably expected to be doing anything. The chaos that could ensue if people were editing... well, it's horrible.

I'm worried about search-engines, though. I don't want the numbers changing all the time. The pages would be unstable, and Google would confuse itself and people searching.

Recalculating: agree, that way lies madness. Unless the numbers were wholly behind the scenes, in which case a main rationale for the numbers is gone.

Numbers also very quickly become confusing for names with more than 5 distinct authors. Steve Martin(43) versus Steve Martin(17)?!

I like Thorold's idea of a qualifying string. Since Tim stated that the split is based upon works, not other author characteristics: any way to make the string work related? The most popular work for that author, perhaps. Though that, too, could require recalculation as popularity changes. So perhaps instead, a canonical choice, with LT deciding upon the canon.

Re numbering authors. With some authors it is easy to decide who is number 1 - e.g. Richard Wagner, John Milton, Charles Dickens*. Often it is not so obvious and, perhaps more importantly, more fluid. When I have split an author I have always put the 'obvious' name (should there be one) at No.1 but otherwise not bothered to try to rank authors with only a handful of copies. The order then may be determined as much by the ease of finding data to separate unknowns as anything else.

Is there anything intrinsically wrong with fixing the numbers as they are? Less room for muddle and I certainly haven't seen very many clearly pre-eminent authors downgraded. As implied in #4, you might get gaps unless the system could automatically close them.

*In passing, Charles Dickens (2) has collaredChristmas Stories - surely not right.

#2 and others, renumbering

I think people have usually done quite a few splits before they settle on a definite method: I know that in the first couple I did when I joined, I put my favourite author as number one. (I think I've fixed all those now!) I can't think of any examples offhand, but I think that's also a pretty regular occurrence for LT authors, who often, naturally enough, put themselves first even when they don't actually have the largest number of books or copies.

#7 The order then may be determined as much by the ease of finding data to separate unknowns as anything else.

That's especially likely with the really numerous splits: the Library of Congress authorities are the most reliable source, so they usually get done first.

With a less than 10-way split, I'd certainly check at the end to make sure that the splits are in descending numerical order, but it's a very finicky job to do that with the bigger split pages.

Yeah, I don't really care if it's perfect. It's just weird when 1-4 are all one-book obscure people, and 5 is a biggie. Worse, I think, is when non-entity self-published guy is #1...

Propsal:
• The numbers are kept as is for the URLs, to ensure link stability.
• When a new split is created, a new number is automatically assigned, i.e. last number + 1.
• Numbers appear in URLs only. Elsewhere, authors are distinguished by most popular work or CK field.
• Authors on disambiguation page are sorted by popularity.

That way, you can put the most popular author on top, and the number doesn't say anything about popularity.

I also like Thorold's idea of a qualifying string - assigning the author birth/death dates or some identifyer to indicate that the main author page has been 'attended to'.

"we frequently get authors calling us up demanding we remove the dates"

Seriously????!!!

Awesome feature, glad you've addressed it.

I'd like to throw in my vote for author birth years beside the names. That would be a much more user-friendly tag for the data. Especially if they're meant to edit CK and put in things like "John Smith (3)" (like for Relationships). I think "John Smith (1910)" is easier. However, what do you do when someone fixes an incorrect year for John Smith to 1911? If that's hard, I guess we're stuck with numbers. Of course, just because you number it a certain way doesn't mean you have to sort it that way. I think I'd rather the authors get permanent numbers and sort 3, 1, 4, 2, 5 than have numbers that constantly shift.

10 is very close to a current proposal in Library land to assign identical names an individual number behind the scenes and then produce the disambiguation information from a separate linked field for viewing in the catalog. (In the library world this would solve the problem of having to replace the name headings on every single book record when someone dies.) We could add a CK field for disambiguation data.

I agree about assigning permanent numbers behind the scenes and sorting and displaying by some other means. Birthdate is good when available.

I'll add my vote behind post #10. The disambiguation page really needs to be sorted by popularity (and maybe optionally by birth date).

Keep in mind the CK bit, though. If numbers are only ever shown in URL, I think it's quite likely someone may fill out the Relationship field as "John Smith (1910)" because that's all they ever see on the screen.

Of course, this just asks for a more touchstone-like feature for this entry where you're actually linking to an author and not just some raw text string.

2>

There's something weird about John Wilson having #1 with only 5 copies of his books, and #2 having 4. Meanwhile 4, 8, and 12 all have over 100 copies. I'm guessing the intent was to order them by time, but I don't absolutely know.

I suspect two factors are at work:

1. People privileging their "own" author as #1, regardless of the number of copies. If one of the authors involved is an LT author, I'd be almost certain that this is what's going on -- who cares if they're the only one with their own books and the other has thousands of copies, they're going to make themselves #1! (LT Author David Mitchell, not the well-known one but one with eight copies of his book on the site, tried this at one point.)

2. Combination. Whether Beowulf appears under Anonymous or Seamus Heaney depends on exactly what the combination situation is at; when it's under Anonymous, it's by far the most popular work there, so the Beowulf poet should be Anonymous #1; when it's under Heaney, all that's left under Anonymous #1 are a few stray copies of Beowulf with some other stuff. Similar factors may be at work for other authors.

My own personal pony, by the way, would be to have separate combine/separate pages for the different split authors. This would be a lifesaver for catchalls like Anonymous; the author combine/separate page frequently simply will not load, and even when it does hunting through for all the various editions of a particular work is difficult and time-consuming. But on the split page it's obvious what needs combining, and if we could just pop over to a combine/separate page for, say, "Anonymous-15", it would be trivial to do the cleanup.

Some difficulties unless/until we get the ability to combine the split authors directly (The examples are hypothetical, but I've seen real ones like this):

Jane Alice Smith is Jane A. Smith (3) and also Jane Smith (7). Do we combine them all into Jane Smith and then split? This goes against the grain of what the combiners have been doing for years. It could also make for very messy author work combination pages.

Similar but worse is if Jane Smith (8) isn't Jane Smith at all, but Jane Smithe. Should these be combined and split out as well in order to get all of the Jane Smithe books on the same page?

>15 jjwilson61: The disambiguation page really needs to be sorted by popularity
I have just done some CK shifting for David Lawrence
http://www.librarything.com/author/lawrencedavid
You can make a reasonable rank order for the first four authors but the next fifteen all have one or two copies each. IMHO that makes ordering difficult, arbitrary and/or meaningless.

For that author name we have a lot of birth dates but don't have one for popular author (2). That limits DoB ordering

David Lawrence includes two different David B Lawrences, a David W, a David Emrys and a David Russell - I haven't looked to see if there are already separate pages for them. The system had pulled in one work by David Lawrence MD and allocated it, apparently at random, to one of the split authors. The David Lawrence MD page was still available separately so I combined it and sorted out the work allocation.

>18 eromsted: I would be inclined not to go down this road until the new feature is bedded down more. I agree that it is irritating to have one author with more than one pages.

#7: I went ahead and reassigned Christmas Stories.

>20 abbottthomas: I agree. But we will need some advice on how to handle this problem. Here's some real examples:

Frank Adams (7) is Frank T. Adams;
Richard Price (11) is actually Roger Price (1).

>22 eromsted: I think your first example is easily fixable - have a look at what I've done. Combined Frank T Adams with the split Frank Adams then edited assignations to combine Frank T's single work with Frank (7). Legal name is Frank T in CK. That seems to me to be OK but easily reversed if anyone disagrees. I am not quite brave enough to give Frank (7) the Canonical name of Frank T Adams.

Your second example is much more difficult - essentially the same as the old problem of combining works where the author name is different and wrong. As both alternatives are split I certainly will not try to combine the two.

>23 abbottthomas:
As Frank Adams is a fairly small case this is probably fine. But note that what you did is essentially the lump everything and then split strategy that I was reluctantly raising in message 18. Before we start doing that wholesale I'd really like to here more from Tim.

>24 eromsted: I think we need some kind of limit to the lumping. It wouldn't be desirable to combine Frank Adams with Adams just because someone entered a work by one of the Franks under that name. I'm not sure if the answer is simply "avoid surname-only combinations", or whether there's a more general principle to be inferred here.

>25 Edward: I am sure that the only circumstance where a surname-only 'author' should be combined is as a temporary measure to allow specific works to be combined appropriately. Once the work combination is done the surname-only author should be separated again.

The problem with doing that is that it can move the pictures and CK around, which then become a lot of work to sort out again.

we frequently get authors calling us up demanding we remove the dates

Okay, not frequently. My guess is that it's once every six months. "Take all my books down" is another one request we get. All told, I feel like I get a crazy author request every month or so.

HNAU's Message (#10)

•Numbers appear in URLs only. Elsewhere, authors are distinguished by most popular work or CK field.

I'm with you in general. Most places they should be either distinguished by that or not distinguished at all. (If some other book recommends "Shop Girl by Steve Martin," and the link goes to the right place, what does it matter if he's Steve Martin (1) technically?).

But I still think we need it on the author page, don't you? If not the individual author pages, at least the disambiguation page?

•Authors on disambiguation page are sorted by popularity.

Yeah, but then you flip to the page where you divvy up the works, and the works are numbered still differently?

The think I like about your proposal is that it goes whole-hog on the idea that authors are distinguished by works and works alone. That's my idea, so I like it. I think, however, you may be tempting me to follow myself too much...

I'd like to throw in my vote for author birth years beside the names.

As a UI element, it's okay, but I can't be the actual answer. That's what libraries do, and it requires all sorts of work, and has all sorts of problems. Someone needs to find out the dates. They even telephone people. Then, in order for the information not to look silly, someone needs to figure out when they died, if the number is starting to make them look over 100. Unfortunately, apart from the biggies, that's hard information to get--impossible, I'd say, in the majority of cases on LibraryThing. All told, it's hard.

So, we could do it as a UI element. But if we're going to do that, we're still going to run into some of the problems above, and we're going to run into the larger problem of doing something very like what libraries do, but not what they do. Cue all sorts of issues, I think.

10 is very close to a current proposal in Library land to assign identical names an individual number behind the scenes and then produce the disambiguation information from a separate linked field for viewing in the catalog. (In the library world this would solve the problem of having to replace the name headings on every single book record when someone dies.) We could add a CK field for disambiguation data.

FWIW, this is a great example of librarian mis-think. The problem isn't the difficulty of replacing headings on books when someone dies. The problem is running a computer system in which such a think is a problem. The problem is trying to run a great information system while ignoring the very concept of relational data!

>20 abbottthomas: I agree. But we will need some advice on how to handle this problem. Here's some real examples:

Frank Adams (7) is Frank T. Adams;

So, my tenative proposal would be to combine them, then divide them (elsewhere described as "lump then split.") The only element that's missing in that case is to have the name for Frank Adams (7) be "Frank T. Adams." (That's a programming issue I will be working on today.)

The problem here is that--at the limit--my advice would result in combining some really silly authors, just to split them out again. The sentiments revolt against that.

Somehow, I think, we've got to turn "Frank Adams (7) into "See Frank T. Adams." That is, allow members to state that a given division of an author should go under another author. The division would still ALSO appear on the disambiguation page. But somehow they would be aliased together. Note that this isn't author combination per se.

Richard Price (11) is actually Roger Price (1).

The paragraph above fixes this problem, but we might also consider some way of allowing members to change—in extremis—not just the canonical name but the canonical identifier for a given work.

I am sure that the only circumstance where a surname-only 'author' should be combined is as a temporary measure to allow specific works to be combined appropriately.

Let's come up with a tool that allows you to do the work without doing stuff like this. Fundamentally, you need a way of comining two work numbers without any shared author, right?

28> you need a way of comining two work numbers without any shared author, right?

Now that does sound good!

But I still think we need it on the author page, don't you? If not the individual author pages, at least the disambiguation page?

I think the global disambiguation notice (for all authors) is mostly needed for transition, until someone separates the data into the individual authors. Most will be put into other CK fields, anyway.

Whatever is still necessary, should be put into the indiviual authors' disambiguation notices, and those should be displayed on the disambiguation page, too. E.g.

John Doe <whatever you put here instead of numbers>
anthropologist, Earl of Doeshire (1905-1981), son of John Albert Doe (1876-1912), also Earl of Doeshire - do not combine (edit)

We still need the numbers for assignment of works to authors (or need to redesign the UI), or to split the original disambiguation notice. I'd propose to put a button/link near the global disambiguation notice to show the numbers.

28>

"Take all my books down" is another one request we get. All told, I feel like I get a crazy author request every month or so.


*boggle* Are they under the misconception that there are illegal downloads available here, or something?

Fundamentally, you need a way of comining two work numbers without any shared author, right?

We have needed that for years. It's such a magical wonder pony that I've never even thought to ask.

28> Yeah, but then you flip to the page where you divvy up the works, and the works are numbered still differently?

Just change the page where you divvy up the works among authors to give them temporary numbers that are the same as the order they currently appear on the disambiguation page. That way the real numbers will remain a URL only thing and you can freely re-order the authors on the disambiguation page.

you need a way of comining two work numbers without any shared author, right?

But, oh the dangerous possibilities... Checks and safeguards? Voting? Scary.

Frank Adams (7) is Frank T. Adams

I think we need to be able to combine them, or at least do the effective equivalent. There shouldn't have to be two author photos, and two sets of CK, and two pages where the same list of works appear, for the same author. It should (at least from the UI) appear to be the same page, reachable from the Frank Adams disambiguation page, or directly from a work whose author is Frank T. Adams.

(And if the system selected author is Frank T. Adams, it shouldn't be possible to assign it to a different Frank Adams from the split page. Actually, in that case it probably shouldn't appear there.)

#28
Fundamentally, you need a way of comining two work numbers without any shared author, right?

That would be wonderful!

>28 timspalding:
Somehow, I think, we've got to turn "Frank Adams (7) into "See Frank T. Adams." That is, allow members to state that a given division of an author should go under another author. The division would still ALSO appear on the disambiguation page. But somehow they would be aliased together. Note that this isn't author combination per se.

I'm not quite sure what you're envisioning here. But the ultimate goal is to have all of the books by an author on one page, no?

I'm reluctant to do any radical combining to make that happen until the system gets worked out more.

And I'm going to keep calling it "splitting" as long as the author page has the text, "If one or more works are by a distinct, homonymous authors, go ahead and split the author." :)

----
Also, any thoughts on the appropriate use of canonical name for the split authors. I think at the moment canonical name isn't working properly here. But when it does, should distinguishing middle names or initials be added to the canonical name, legal name, or other names?

Fundamentally, you need a way of comining two work numbers without any shared author, right?

There is already a URL that will do this. There just needs to be UI to get there, and some better confirmation checks built into the UI after hitting that URL.

Whether or not the numbers are recalculated, it would be nice if we could get the author with the most books at the top of the page, even if that meant John Doe (4) was listed before John Doe (1) and John Doe (2).

>28 timspalding:

Okay, not frequently. My guess is that it's once every six months. "Take all my books down" is another one request we get. All told, I feel like I get a crazy author request every month or so.

>31 lorax:

*boggle* Are they under the misconception that there are illegal downloads available here, or something?

Somehow I get the feeling this is part of the same mentality that led to some independent booksellers demanding that their event information not be syndicated across the web.

And to actually stick with the main topic of this thread, just want to point out that even if authors are ordered by number of books, there's always the possibility of a new author appearing with a burst a popularity that catapults them from 10th place to 1st in terms of copies; this is especially prevalent with celebrities who write one book but share the same name as a lesser-known author with 30 books to their name. In an instant that established author drops from #1. We don't really have a way to keep up with that.

#38: Or people like Steve Jackson (1) and (2), both being living authors, passing each other. Especially many books by Steve Jackson (US) are actually by his company and have real authors they could be attributed to; if they move in and out, like Beowulf, they could easily switch places repeatedly.

On the numbering vs. birth and death dates, can I make a suggestion?

Can we use the CK birth and death dates for display purposes, if not for data distinguishing purposes?

So, could we have the top of one of the divided pages say:

John Doe (1752-1818)
then John Doe (5) in smaller letters beside it or under it?

or
John Doe (1935-)
John Doe (2)

Ditto for the list of names on the disambiguation page; I think it would really help to clarify things for the average user.

38> And to actually stick with the main topic of this thread, just want to point out that even if authors are ordered by number of books, there's always the possibility of a new author appearing with a burst a popularity that catapults them from 10th place to 1st in terms of copies; this is especially prevalent with celebrities who write one book but share the same name as a lesser-known author with 30 books to their name. In an instant that established author drops from #1. We don't really have a way to keep up with that.

That's not such a problem. Just have LT order the authors on the disambiguation page by number of works as it is when the page is rendered. If they then want to assign books to authors, have that page number the authors by the order they appear on the main page, not the number in the URL.

*boggle* Are they under the misconception that there are illegal downloads available here, or something?

That's the most common reason. But I just had this crazy French author who asserted her publishers didn't pay he royalties, so we couldn't list the books for sale. When I explained to her that LT wasn't a store, she told us she was going to sue us. It was actually a rather fun exchange. I don't suffer fools, after the first few emails anyway.

Need to reply from 31 on. Much to say.

Tim,

Harking back to the OP, the fundamental entity is a member's book, or rather 'published work'. Generally, that is linked to an author or 'primary nominated producer'. That single link between the work and the identified 'copyright holder' (for want of better word or words) is what everything else hangs off. The problem is that there are sometimes many authors for a work (anthologies; graphic novels) and if you're moving towards a sophisticated author system, then you shouldn't just be thinking of many (works) to one (author), but many to many relationships. (Perhaps the 'other authors' fields finally get brought into play.)

Now the 'author' isn't a name, it's an entity of which one form of attribute is name. And name can be legal name, changed name, acquired name (marriage, title), role, publishing house, pen name, 'joint name' (where the individual authors - such as wordsmith and illustrator for comic books - aren't separately identified). So, if the work is only going to be linked to a single entity (which is what you're struggling to achieve) then that entity has to be an abstraction to which any number of 'authors' (or author names, or 'associated attribute sets') can be linked. Sticking with the names or name strings as the sole entities that you are trying to link/manipulate (and from which you suspend the characteristics/attributes such as dates and gender) becomes a nightmare.

I think you are going to have to start building many to many capabilities in, with the target/s for the work being 'authorial entity/ies' identified by a number. For UI purposes, a number is useless, so it's going to have to be a canonical name. Linking to that 'canonical name' are going to be 'known names' (legal names; or names used by the LoC for instance) and 'entered names' (which include misspellings; suffixes like MD; surname-forename).

Now, what I've described is an enormous leap. Having started down the path, I don't think you can stop now but you won't get there in one go. All I can recommend is that you bear in mind what the end-point is before taking any further steps.

Would it be possible to use this as a springboard to liberate yourself from the string limitation of authors? What I mean is this: could you take the lnamefname-# page of each disambiguated author, and then make it a redirect to (e.g.) author/####### for some number. Then, authors can be merged and split like books.

And for future authors, the system could treat each author-work combination similarly to the edition-level of a book, auto-merging the author if the book is automerged with another, and the author has a similar name.

Let me try an example: Say, for the sake of argument, there were two Isaac Asimovs. One wrote popular science fiction (and just about everything else), and the other wrote gardening guides. If you went to the page for asimovisaac, you'd get the disambig page, like now. And if you clicked on Isaac Asimov (1) (say, the more popular one), you'd get redirected from asimovisaac-1 to author/1234567, whereas, if you clicked on (2) you'd get redirected to author/0987654. If I add a copy of Foundation to my library, it'll get merged with the existing copies of Foundation, which will then merge that Isaac Asimov with #1. If I add a copy of Extreme Mulch, it'll get merged with other copies of Extreme Mulch, which will in turn merge that Isaac Asimov with (2). If I add a third book, say, Nautical Expressions of Middle Europe by Isaac Asimov, assuming I'm the first person to put it up, the system would assume that it wasn't THE Isaac Asimov, or the other one, unless somebody went in their and said so. Or, maybe if there were some heuristic to say: "this book has a lot of tags in common with this author, maybe I should automerge it with him until someone tells me otherwise."

Just a suggestion.

Have y'all stopped yer philosopizin'? Trouble a-brewin' over yonder.