My Member Recommendations list is growing unwieldy, and it's a pain to have to slog through pages and pages to find the new ones. How can I delete the ones I've already seen?

It would be great if you could check a box next to the recommended title and then delete all selected ones with just one click. In fact, I would love that feature for the Automatic Recommendations too.

There already is such a function in automatic recommendations: the "No thanks" link.

Having the same thing appear on Member Recs has been asked for several times already.

2>

It's enormously frustrating that Tim hasn't been willing to entertain a "no thanks" for member recommendations.

For someone with such a wealth of data available in the form of library similarities and so forth, he seems strangely unwilling to actually use the data, insisting instead that human-generated recommendations are always more reliable than any automatically-generated recommendations -- you could do a much more robust recommendation system by looking at clusters of books, rather than individual books, and, yes, by considering ratings. (His insistence that ratings are meaningless because they skew high is nonsensical; if the average rating is 3.8 with a standard deviation of 0.2, to invent numbers, doesn't that mean that my ratings of half a star are extremely statistically significant and should be taken into account, at least when generating recommendations for me?)

I'd also like to add my frustration that he refuses even to collect data about what books people have actually read, despite claiming that (at least when it comes to movies) this is more useful for generating recommendations.

correction: "most useful", not "most useful"

Message edited by its author, Oct 21, 2009, 1:10pm.

4>

I think he thinks the default assumption is "read". I personally would be happy with a "if it's not marked as unread, it's read", since there's already the structure for that; I know that this doesn't properly handle dictionaries, and that you have a more stringent definition of "read" than many people, but I think it would be 90% of what you want for a tiny fraction of the work.

I think he thinks the default assumption is "read"

No, he thinks it demeans the book (and maybe the whole process of reading) to say that you've "read" it; our relationships with books are ongoing journeys that are never really complete.

a tiny fraction of the work

What work? It would be incredibly easy to add an Already Read collection.

>4,5

That was from a discussion of the Netflix million-dollar algorithm and the developers. One discussion that stated, 'having watched the movie was more of an indicator of future movie viewing likes than the starred rating on past movies.' It was on the thread that said stars should hold weight in your recommendations.

I don't think Tim's ever said that he's assumed that all the books are read. Far from it - he's always stated that for one to possess a book, one has an affinity for it in one way or another. Nothing more, nothing less. Basically, Tim assumes that those that log books here aren't just grabbing random books from a stranger's house while blindfolded and putting them in accounts.

>7 Okay, edited to "more useful" from "most useful". But I still maintain that if there's some useful data that plenty of members would readily volunteer if only they had the chance, they should just give us the chance to provide that data.

Basically, Tim assumes that those that log books here aren't just grabbing random books from a stranger's house while blindfolded and putting them in accounts.

Apparently he's never been to one of those fill-a-bag sales.

8>

Well, I've been to fill-a-bag sales, and I certainly don't grab books at random. YMMV.

6>

As a default collection that everyone shares? I don't know that. But there already is an unread collection.

Personally I wouldn't use an Already Read collection, since I see it as an insult to my intelligence. I have clearly marked unread books, and have no need to enter the same information in another form out of concern that someone might otherwise think I have read every page of a dictionary.

>8 I've been to fill-a-bag sales and I certainly don't fill up my bag with just anything that hits my hands. That's a waste to me and entirely unfair to other people who may very well want to read books that I could have just greedily shoved in a bag "because it fit."

Everything I throw into that bag is something that catches my eye for one reason or another. But if I show up to a bag sale and it's all books on Camping, How to Buy a New Condo and Raising Chinchillas for Fun and Profit - well, I'm passing.

>9 I think your intelligence is far too easily insulted. Making assumptions based on lack of data just isn't generally effective. The fact that you mark unread books doesn't mean that everyone does; some people don't specify either way, some people only use the TBR collection for books that they're meaning to read in the near future, etc. Site-wide inferences based on your personal use of the site would be a lot less accurate than directly-supplied data. That's not a personal insult, just a basic fact.

Everything I throw into that bag is something that catches my eye for one reason or another.

My point was that the collection of books that "caught your eye" isn't nearly as indicative of your reading tastes as the collection of books that you've actually devoted hours of your time to reading.

Yes, but caught my eye is still indicative of some sort of desire to spend time/money/something.

You want optimized data in heuristic world. That doesn't happen.

Sorry, how does the impossibility of perfect data imply the impossibility of anything better than the status quo?

Two further thoughts:

How were you planning to implement the "read" aspect of your lists? Do you think it would be best to base it on whether people have catalogued the book but not marked it TBR?

And adding a new way of getting recommendations doesn't have to eliminate the old way. There are several different book-to-book recommendation lists, and people can look at the ones they find most useful.

11>

I'm not talking about "everyone's" data, I'm talking about mine. Look, I have a pile of books marked as "unread". It's insulting to have someone, or some algorithm, go through each and every book I have that isn't so marked and pretend they have no idea whatsoever about whether I've read it or not. No, it's not going to be 100%. But it's pretty damn close, and your insistence that the existence of an "unread" collection or tag in someone's library doesn't have any bearing on whether people have read books that aren't in that collection or tagged that way is just baffling. I think you are smart enough to figure that out -- if you don't think I'm smart enough to do the same, that's an insult to my intelligence. Obviously, if someone doesn't use To Read or have anything tagged "unread", I don't know whether they've read their books or not, but that isn't what I was talking about.

No, it's not going to be 100%. But it's pretty damn close

I think we have vastly differing definitions of what "pretty damn close" means. I've catalogued 2,242 books; of these, 552 are tagged "read" and 762 are tagged "unread".

LT isn't restricted to light fiction and narrative non-fiction. It's pretty disingenuous to pretend that the only reference books people have are a few dictionaries.

Even if there were no difference, though, I don't see how it's an insult to your intelligence if the computer requires input in a certain form. It's like putting your ratings as a comment or tag ("5 stars") and complaining that they're not included with the ratings on the Work page.

If one person wants to enter ratings as tags and no other way, does that mean no one else should be allowed to have a rating field?

16>

All I meant was that it's pretty damn close in my library.

(And perhaps we have different definitions of "narrative non-fiction", but I wouldn't say most of my non-fiction, which I have read, is "narrative".)

I'm not necessarily opposed to some sort of "read" indicator, but another collection that for me, personally would be redundant is not useful for me, personally.

I looked at your library, excluding books tagged either "read" or "unread", and while there are certainly some reference books there (The Elements of Style leaps out at me), in addition to your dictionaries and textbooks, it's hard for me to understand how something like the Aeneid is something that intrinsically is not suited to being described as either "read" or "unread". At any rate, though, I didn't mean to speak generally, and I apologize for my imprecision.

I personally would not find a "Read" collection to be useful, and I objected to the implications (as I perceived them) that nobody could possibly get useful data from their own library by assuming that books were read unless indicated otherwise. I didn't mean to say that either of these were the case universally.

Oh, we were talking a cross-purposes. I was talking about the utility of the feature for the site as a whole; personally, I'm happy with my tags and don't need a Read collection, but I think it would be useful for aggregating data across users.

(And the necessary clarification, after all the review-comments discussion: even though it would be redundant, I wouldn't mind copying my books into it if it actually did something. I'm not advocating a feature that I want other people to have for my benefit but wouldn't want myself.)

A lot of my classical texts, like the Aeneid, consist of maybe 30 pages of the actual text followed by 200 pages of commentary, which I only look at selectively.

Of course, my tagging may not be perfect, and there are times when I actually can't remember whether I've read a book.

You're misinterpreting what I'm saying.

You want to apply standards that would work in optimized data environments. We will never have it - not even close. If development were to go down that path, you'd be sorely disappointed. You often like to say, "if we develop something that would use X piece of data, X piece of data would get better." But you actually have no proof of concept on that - even memes that tie into CK data only get momentary spikes and then level out, they don't get sustained usage.

Heuristic analysis is different, but incredibly valid in the face of non-optimized data. In fact, it's much better than trying to cram the square peg of heuristic data through the round hole of optimized analysis hoping that somehow either the data will become optimized or that the analysis will cause optimized data to appear.

I never said not to do anything, I'm just pointing out that you're rooting through the wrong toolkit - you want to use a slide rule and the data set really calls for a salad spinner.

Oh - and on the read feature of lists - that's a whole different beast and it's more of a 0,1 in terms of database-speak and would likely be implemented via a single click. It could even go so far as a 0,1,2 where 2 could assign it only to the standing "to read" collection (but this is a huge maybe based on the next paragraph).

But all it really does is give the list the ability to come up with a percentage of "completeness" - it would have no real bearing on your catalog. Because you should be able to mark something on a list you've adopted as "read" even though you don't have it in your library at all. For instance, I don't currently own a copy of Lord of the Flies, but I know that I've read it - multiple times. I also don't own a stand-alone copy of Of Mice and Men (my copy is coupled with Tortilla Flat. And yet, both of those appear on the 1001 Books List. I'd want to mark them as read without having to put them in my library since I'm not planning on re-reading them and I'm not planning on buying new copies anytime soon.

That's a complete derailment for now - but the "read" really would be a list functionality and not so much a catalog functionality. I probably should have called it "complete."

Either way, it's a simple matter of using that "mark" to say %C=R/(R+U+TBR)

even memes that tie into CK data only get momentary spikes and then level out, they don't get sustained usage.

This is because once the data has been entered once, it's done. If an author-gender meme is added and people go crazy entering author genders, it's obviously not going to be sustained because we don't need to add more author genders every day. The same thing would apply for ratings or a Read collection: a new feature would attract interest, and people would add some more ratings or transfer books to the Read collection immediately. Once they had done that for their main collection, it would of necessity slow down to their reading pace.

I'll freely admit that I know nothing about data analysis, but I still think there's a basic common-sense point to be made here. People are more interested in entering data that does something. If a new feature is added, they'll enter more of the relevant data. This obviously won't be sustained, since there's only so much data to be entered. But the extra data from the initial spike will be there, increasing the value of that data set overall.

I think the CK memes were a success and did increase the value of CK. Do you not agree with this?

Finally, I think you often dismiss the general interest factor of things like this. Even if the data doesn't accomplish any greater purpose, it would still be nice to have an easy way to see who had read a book.

Because you should be able to mark something on a list you've adopted as "read" even though you don't have it in your library at all.

I think this is the whole point of Collections. The Read collection would be distinct from your library (I hope there will eventually be generic editions). I can't imagine that they would go to the trouble of implementing a non-collections based Read feature, since all book data is catalogue-based.

If you're saying that the Read data would be list-based instead, does that mean that I'd have to go through each list separately checking off the books I'd read, since there would be no independent record of my reading separate from the lists?

I think the CK memes were a success and did increase the value of CK. Do you not agree with this?

I agree that it was fun for a day or two for some people. I run across authors without sexes every day - with plenty of books owned. I run across many more with little to no birthdates filled in. It looks something like this in rough CK entry land:

~~~~~~~~~/ ~~~~~~~~

So you have a spike - some data gets filled in, but not all and not by a long shot. Not even the majority. What happens is that some active users who don't like seeing incomplete stuff like to see things filled in for themselves.

And saying data that does something is another fallacy. Meme data exists for awards, for honors, for series, for locations, for characters and rather spiffy clouds exist for events and blurbers. And these have experienced minor spikes, but since they require more effort, folks tend not to sustain the spikes as long.

Entering data is work. There's no easy answer to it. And data that touches catalogs becomes personal.

I don't dismiss your interest - I've done data analysis for several years. There's an underlying assumption that data can do everything and anything - and that it exists. Another analyst and I used to joke about the ODBC driver of everything magical in the sky that we'd connect to for a certain manager.

This was a funny joke between us...I promise.

I will concede one secret - given enough money and time, one can analyze anything. Heck, the human genome only took 3 billion dollars and 20 years. But this time and money isn't available or practical in most situations.

Which is where heuristic analysis comes in...I'd use this mostly in Risk analysis after the fact when I was asked to prove something like "how many repossessions did implementing X or increasing Y" avoid? This is a dicey question - because I really didn't have time to do that type of hard-core analysis. If I did, I would have been doing economic analysis, I would have been analyzing past repossessions and ensuring that I'd normalized all geographic, economic and whatever statistical minutia I could think of along the way. I could have literally spent MONTHS doing this type of analysis (and would have needed a much higher degree and far more maths than I was comfortable using on a day-to-day basis). Heck, I would have been normalizing the auction prices to ensure that I hadn't sent one car to auction sooner in either sample...well, you get the point. Optimizing the data - while not impossible per se - way too costly and too much time involved for a simple question.

Instead, taking a heuristic approach means paring down the data and taking a rule-of-thumb-type approach. I have to basically stop worrying about the metric-ton of data I could worry about and focus on a limited set that will make sense and produce results from that analysis. And the truth is - it's a very reliable analysis of whether or not we're doing the right thing. The time/money that would be spent on the hardcore analysis might get us that extra feel-good +/-X%, but at what cost? If the cost is a month of analysis and pulling in an additional expert in economic analysis that will cost the company $20,000 dollars - well, that's not worth it.

And this is the problem that we run into with read dates, stars, reviews, currently reading, tags, collections, multiple collections, and any number of extra pieces of data. It sounds easy when you say it - it always does.

When it comes to the algorithm - the likelihood of new data fixing things is pretty slim. What would probably fix things would be digging into the questions that are being asked.

After all, Tim's base supposition is that "you are attached to your books" - there are those here that feel this supposition is wrong. And yet, what have we really based this on? The occasional, "OMG I would never read The Secret Life of Bees!" comment.

If Tim had the time, a random survey of a lot of recommendations over a few weeks with 4-6 pointed questions (2-3 things about the book that recommended the work and about the work as well) that would help support/disprove these assumptions would probably help a lot and lead to better tweaks in the algorithm.

If you're saying that the Read data would be list-based instead, does that mean that I'd have to go through each list separately checking off the books I'd read, since there would be no independent record of my reading separate from the lists?

The lists would be attached to a work record. So once you've marked a work as read, it's read across all lists. I imagine you'd probably have to check off everything at least one time through. I don't know that it would be attached to anything else. People seem really weirded out anytime anyone suggests anything automatic in relation to reading dates.

Thanks for explaining this further.

So you have a spike - some data gets filled in, but not all and not by a long shot.

Right, but there's still more data than there was initially.

Entering data is work. There's no easy answer to it. And data that touches catalogs becomes personal.

Right, and people are more likely to care about personal data. Tim has made this point before when comparing LT tags to Amazon tags.

Also, at least for Read, the work involved is pretty minimal. Lots of people do much more work every time they read a book: rating, review, reading date, remove from CR, update multiple threads, etc. I don't see why Read would be used any less than CR; in fact, I think there would be much more Read data than CR data, since CR data is only temporary.

I have to basically stop worrying about the metric-ton of data I could worry about and focus on a limited set that will make sense and produce results from that analysis.

This comes back to the point I was trying to make about data that "does something" vs. data that's just there. Gathering Read data doesn't automatically require that you analyze it; I think it would still be worthwhile even if it weren't incorporated into any algorithms or analyzed in any way, but just showed up on the Work pages. Collecting the data is easy. Displaying the data is easy too. The hard part only comes when you try to find some deep meaning behind it... and sometimes, that just doesn't matter. Tim could take Read data and put it into the same algorithm that he already uses for books catalogued in general. Maybe the results would be "good", maybe not. But they would be interesting regardless.

The lists would be attached to a work record. So once you've marked a work as read, it's read across all lists.

So if the only list containing a particular work were deleted, there would no longer be any record that I had read that book next time a list was created?

I still think it seems harder than keeping track of the read books via individual catalogues, though I can't entirely say why. With Collections, it would be very straightforward; I'm still not entirely clear on how your proposed connection between individuals and works and lists will work.

People seem really weirded out anytime anyone suggests anything automatic in relation to reading dates.

I think this is only when something automatically happens to the reading dates; I don't remember any objections to the idea of using reading dates to populate a Recently Read collection, as long as that collection could also be used without reading dates.

I think your boundless optimism in getting even a statistically significant number of people to use "read" in a decent fashion so that an algorithm could process their catalog to decently use it in a recommendation algorithm is nice.

Nice, but impractical. For it to work, it would have to be complete for a person's catalog. And Tim would eventually have to develop at least two algorithms - one that took those that used read versions in a statistically significant manner and one that did not - and that would be a lot of processing for little to no gain. You're convinced that there would be a ton of gain, and yet you don't give the heuristic methodology any credit when so much of the world actually operates on this very principle.

So if the only list containing a particular work were deleted, there would no longer be any record that I had read that book next time a list was created?

Again, you're attempting to co-opt one thing not meant for another. But, in the public listofbests world, once a public list is adopted by someone other than the originator, you can't simply delete the list. You can un-adopt it and never see it again on your list, but as part of someone building a community, they don't give you the power to tear down the community just because you felt like clicking "delete" one day.

To be honest, I believe there are two sock accounts that deliberately adopt every substantial list they can for this very reason - so that the community never loses decent lists.

You're convinced that there would be a ton of gain

No, you put these words in my mouth. I said it would be interesting.

For it to work, it would have to be complete for a person's catalog.

This isn't true. It might even be better if it were incomplete, based on more recent reading rather than every children's book someone read in the distant past.

And LT's basic recommendations "worked" when there was a lot less data than they had now. Why is it that anything less is always assumed to be totally insufficient?

But again, what does "work" mean? I'm not sure how you get from "it would be interesting to gather this information and see what it does" to statements about how we don't know whether it would "work" or "lead to a ton of gain" or whatever. We're not dealing with a life-or-death decision here; it doesn't matter how effective it is, it could still be fun and interesting.

And Tim would eventually have to develop at least two algorithms - one that took those that used read versions in a statistically significant manner and one that did not

Does he do this now, distinguishing between those who have catalogued in a "statistically significant" manner and those who haven't?

But, in the public listofbests world, once a public list is adopted by someone other than the originator, you can't simply delete the list.

Thanks for this clarification.

Again, you're attempting to co-opt one thing not meant for another.

I really don't see it this way. It seems to me like you're trying to ignore the most obvious tool for what you want to accomplish. Why re-invent the wheel? It would be a waste of developer time to create an entirely separate mechanism for storing member reading data related to lists, when they could easily use what they already have to store reading data in general and then apply it to the particular situation of lists.

It would be a waste of developer time to create an entirely separate mechanism for storing member reading data related to lists, when they could easily use what they already have to store reading data in general and then apply it to the particular situation of lists.

Only if you wish to co-opt the read data for something else. If the list functionality exists to serve a specific purpose, it's not a waste of developer time and it produces a ton of data to use both personally and for individuals.

It just doesn't work in a way that you'd like to co-opt.

If it did work in your co-opt way, here's the main problem:

How do I mark a book as read on a list, but not mark a book as read in my library for recommendations? I read book XYZ just for this list challenge and I have no desire to have this show up in my read algorithm. I only want the list to show my progress on the list.

You really should develop a survey and ask Tim if he would implement it - see how many people really want said read vs. catalog approach. I believe you're far too wrapped up in "wouldn't this be interesting?" to see the amount of work vs. what would get done. You see this:

there's still more data than there was initially

As proof of success in the "data that does something" argument. And yet you still have no proof of concept that individuals would really meticulously go through their catalogs and use your feature as you intend. You are on a board where the users are by-in-large the same 200-300 users that know the bulk of the features.

If you don't have good surveying experience and Tim is open to it, he may work with you to write up a survey that would help you see not just the idea of interest, but the level of effort individuals would likely put forth. For instance, I just went through 30 people that had posted - I didn't find one collection that indicated "read," and I found about 3 people that weren't using collections at all. This is amongst the more-knowledgeable users (everyone also had more than 200 books in their library). I just find it interesting that individuals aren't naturally selecting this collection - and yet some form of "these books sucked and I couldn't finish them" collections appeared with 5 users.

Anyway - you really need proof of concept to show the following:

- Affinity habits
- Time spent on catalog
- Time spent looking at recommendations
- Recommendation satisfaction
- A few questions regarding recording read books

You'd probably want about 10 questions and you'd want them to be normalized so that they don't read, "if having a read collection would lead to really awesome recommendations, would you spend 10 minutes updating your catalog with a really simple fix to do it?" You'd want some baseline questions like, "How much time a week do you currently spend updating your catalog information (adding books/updating tags/updating collections, etc.)" - with time options.

Oh - and you'd probably want at least one open-ended question to get verbatims - those often reveal a lot more than the questions.

Only if you wish to co-opt the read data for something else. If the list functionality exists to serve a specific purpose, it's not a waste of developer time and it produces a ton of data to use both personally and for individuals.

How is this "co-opting" the read data? Isn't using Read data to show which books people have read sort of the point?

What is the specific purpose that a separate list-based Read functionality would serve? How would it produce a ton of data that wouldn't be there if it were stored in a Read collection? I think the list UI would be pretty much identical either way.

How do I mark a book as read on a list, but not mark a book as read in my library for recommendations?

This is why there are separate collections. Your Library is not the same as the books you've read. If you want to know how LT can figure out that you've read a book when you're not willing to admit it, you're out of luck.

I read book XYZ just for this list challenge and I have no desire to have this show up in my read algorithm.

How does a book that you've only read for a specific purpose not count as read? You've still read it regardless of the reason.

You really should develop a survey and ask Tim if he would implement it - see how many people really want said read vs. catalog approach.

If Tim showed any interest in the feature, maybe. But I don't think he particularly cares, especially if there are vehement arguments that the simplest approach isn't good enough.

It makes no sense for my personal cataloguing to separate out childhood series from My Library. I still own those books, I just don't want recommendations based on them. But Collections are a convenient way to determine what books are included in recommendations, so that's how it works. One of the main justifications for all the time spent on Collections was that it provided an easy way to assign meaning to a set of books. They're not going to develop a new way to assign meaning to a set of books just because you want the meaning without having to acknowledge the books.

I believe you're far too wrapped up in "wouldn't this be interesting?" to see the amount of work vs. what would get done.

No, the list issue isn't about "wouldn't it be interesting?" It's about already having a method in place to do what you want: store a record of the books people have read, to be used in this case for calculating list completion. Why would they not use the tools they already have?

For instance, I just went through 30 people that had posted - I didn't find one collection that indicated "read," and I found about 3 people that weren't using collections at all.

Do you have a list of these people (or the thread you took them from), so that I can look at the same ones you did?

The fact that people don't currently have a Read collection doesn't mean they wouldn't use it. It would have been more helpful to check how many were using Currently Reading or To Read, similar sorts of default collection, and how many were using a Read or Unread tag--in other words, how many people are recording this sort of information one way or another. I don't even have a Read collection, because collections don't do anything different from tags. That doesn't mean I'm not keeping track of that information on LT.

I think a survey would be interesting, but complete overkill for something like this. Rather than spending hours analyzing whether people would use a default Read collection, it would be much easier just to try it out. Maybe the process would end there, or maybe it would get enough use that Tim would want to try using it to generate recommendations. But there's no point in spending more time analyzing the possible outcomes than it would take just to do it.

The fact that people don't currently have a Read collection doesn't mean they wouldn't use it. It would have been more helpful to check how many were using Currently Reading or To Read, similar sorts of default collection, and how many were using a Read or Unread tag--in other words, how many people are recording this sort of information one way or another.

Well, for some of us, the fact that we do track this information now (using Currently Reading, To Read, and an "unread" tag) -- is exactly why we wouldn't use a redundant-to-us additional mechanism of representing the same information.

>29 But would it still be redundant if it did something that the other methods didn't? I don't have a Read collection now, but if there were a pre-defined Read collection that did something as simple as showing up on the work pages like Wishlist or CR (or like they used to), then I would take the two minutes to copy my read-tagged books into the Read collection.

30>

Maybe if it did something that was useful to me, which your example wouldn't. Even then, if it was a collection it would have to be very useful, since on the whole I try to be minimalist with my collections.

Zoe, it's co-opting much in the same way that saying, "hey, let's take currently reading and convert this to 'needs to be reviewed' collections!" would be a co-opt. They seem naturally related, but they're really not. One does not lead to the other.

As to this:

If Tim showed any interest in the feature, maybe. But I don't think he particularly cares, especially if there are vehement arguments that the simplest approach isn't good enough.

But you're willing to post on and on that you're sure this all works based on what? No data, no proof of concept, no similar workings and nothing other than "but I want it?" PROVE to Tim that you want to take this to the next level.

Pardon my French, but either put up or shut up. You're stuck on, "but it's interesting" and "hey, this sort of looks like something I want so I'll make this square peg fit my gimme!" instead of actually taking the reigns.

Tim has his data and a heuristic model that works. You believe you know more and that you have an even better way of doing things. SHOW him that you're willing to get empirical data that shows you're serious and can prove that users will behave the way that you believe they will. Right now, there's a lot of world data that says, "people don't fill in things." You believe LT users behave differently - show Tim you want to prove it. If he won't let you, use a poll site that will let you.

Do you have a list of these people (or the thread you took them from), so that I can look at the same ones you did?

I looked at my groups view and looked at 30 unique names on the front page of people who had last posted in the 1-2 hours before I posted that message to you - it was fairly random. That's how you survey...

Zoe, it's co-opting much in the same way that saying, "hey, let's take currently reading and convert this to 'needs to be reviewed' collections!" would be a co-opt.

Except that "currently reading" and "needs to be reviewed" are two different things, while "read" and "read" are the same.

But you're willing to post on and on that you're sure this all works based on what?... PROVE to Tim that you want to take this to the next level.

What's "all this"? Adding a Read collection isn't hard. Why is it that any of my suggestions need to be subjected to detailed analysis before they're even admitted for consideration, while this doesn't apply in your case?

Right now, there's a lot of world data that says, "people don't fill in things."

Again, why does this apply only for me, even for simple things like adding an extra default collection, but not to your list proposal?

Anyway, people who barely use the site don't fill in things. If you look at a sample of active users, like you did, you see that 90% of them are using collections--"filling things in". I only looked at half the number you did (due to computer issues and also the obviousness of the conclusions), but of 16 people selected in the same way, 9 were using the CR collection, 11 were using To Read, 8 were using Wishlist, and 6 were using Read but Unowned. One was using no collections at all. Seven were using no user-defined collections (including the one with no collections at all).

This just tells us what we already know: the defaults get used, and people don't do much beyond the defaults. Though there were a few people with a lot of user-defined collections, the median number of user-defined collections was 1.

So, the pre-set collections get plenty of use.
There are hardly any user-defined collections.
There are hardly any user-defined Read collections (I did actually see one).

Do you really think the best conclusion to draw from this is that people aren't interested in a Read collection, rather than that people just tend to stick to the defaults?

Message edited by its author, Oct 22, 2009, 7:39pm.

32> Tim has never performed a survey before deciding whether to implement a feature (for good reason). I think it's completely ridiculous that you would hold Zoe to a higher standard.

Message edited by its author, Oct 22, 2009, 7:50pm.

>34 And you obviously haven't read the bulk of this discussion. And you're wrong - Tim's asked for input plenty before implementing a feature. And this survey isn't about whether the feature should be implemented - it's to show that there's enough on the part of the user base to behave differently than would be expected.

Tim has defended his heuristic approach to recommendations numerous times. Zoe would like to change it, add a read collection and base recommendations off of that because her hypothesis is "if the data did something, people would use it."

In a random sample, there were no read collections, but there were abandoned collections - it could be a fluke, it could be that people don't care to segregate. We have proof that data that does things results in short bursts of input, but on-going data that does things absolutely languishes. Real-life data that relies on continued input requires either an in-your-face interface that forces all users to conform to the data or it tends to fail - not just in LT, but across the universe of data collection in general (there are universal rules on how to accomplish these kinds of things). The heuristic analysis approach never ever says, "take incomplete data because it seems more interesting" - and yet, this is exactly what Zoe proposes.

Zoe is optimistic still feels the LT users would perform differently.

Tim won't change his mind without proof at this point - the history of this type of analysis is simply not on her side. The history of people and doing the work required to do this is not on her side. And yet, she sees something in the users here that tells her that a large enough population will behave differently enough that Tim should entertain it.

It's not a higher standard - it's the one sane chance of convincing Tim that he should really entertain it.

But thank you for reminding me why it's a waste of time to bother explaining how analysis works in the face of complex data or even bothering with why heuristics works. We should just continue with having people type endlessly and getting frustrated and taking things personally when Tim doesn't implement things.

Or, they could take the initiative, and find out that either 1) they have support for their idea and show it to Tim or 2) they could learn that those that have developed heuristic analysis methodologies did so for a reason.

Do you really think the best conclusion to draw from this is that people aren't interested in a Read collection, rather than that people just tend to stick to the defaults?

You proposed a while ago when this first started that this was a vital collection. What was that based on? Other than you found it interesting?

Because you're back to your theory which is "if Tim would just build it, they'll use it." And you still need to find the proof of concept.

ETA - I'm not saying this to be mean. There are tons of things that are interesting in this world. There could be a huge case to make "unreviewed" a default collection if Tim wanted to be self-serving. After all, it benefits Library Thing for Libraries, it could result in more reviews on the site and make for more user-driven data. But we don't do that.

Other people might like default "Gift" collections because they want to know what the most received gift books are - maybe I'd like to see Tim recognize challenge groups and have him create a Challenge collection so there could be a Challenge module on the home page where folks could see books that have been read in challenges and challenge groups could be highlighted to drive traffic to those groups.

It's all interesting and it can all serve a purpose. But there has to be something other than "it's interesting."

Honestly, I'm giving you a pretty good way to go to get Tim's attention. I'm not kidding.

Message edited by its author, Oct 22, 2009, 9:02pm.

>34 Thanks for the support, yet again.

>35 And you obviously haven't read the bulk of this discussion. And you're wrong - Tim's asked for input plenty before implementing a feature.

What you're insisting I do goes far beyond "asking for input". The input Tim has asked for before has always come in the form of Talk threads, but suddenly that's not good enough here.

There's an incredibly easy way to find out whether people would use a Read collection: implement it and see.

Tim has defended his heuristic approach to recommendations numerous times. Zoe would like to change it, add a read collection and base recommendations off of that because her hypothesis is "if the data did something, people would use it."

I don't think you've read what I wrote earlier. I don't think Tim needs to replace the current recommendations algorithm; I do think there's more than one way to get recommendations, as you can see on any work page. And a Read collection could be used for more than one thing (including your Lists feature); recommendations are just one possibility.

In a random sample, there were no read collections, but there were abandoned collections - it could be a fluke, it could be that people don't care to segregate.

And in another random sample, there was a Read collection. Considering that the median number of user-defined collections seems to be around 1, there's not nearly enough data in your sample to draw reliable conclusions about the frequency of specific user-defined collections.

Fortunately, that doesn't even matter. All of the pre-defined collections have plenty of use. The safest conclusion is that if a collection is there, people will use it (at least, as long as it's a reasonable collection--it would be interesting to see what would happen with a randomly-named pre-defined collection).

Real-life data that relies on continued input requires either an in-your-face interface that forces all users to conform to the data or it tends to fail

So why do we have pre-defined collections at all, if they're destined to failure? You constantly say that I'm expecting something anomalous here, but what makes you think that Read would be different from all the other pre-defined collections? I'm expecting the same behaviour that's been seen in the most closely-analogous data set we have; you're the one who thinks that this situation is completely extraordinary.

You proposed a while ago when this first started that this was a vital collection. What was that based on?

Do you remember the context of this? I can't see myself saying something like this is "vital".

I'd imagine my main argument would have been logical consistency. The current default collections are a pretty random assortment; rather than one particular combination like "Read but unowned" (which I think was the least-frequently used in my mini-survey), it would be a lot cleaner to have separate collections for Read, Unread, Owned, and Unowned, and to allowed us to look at the intersection of different collections. There's no good reason for "Read but unowned" to be privileged above all other states.

And then we have To Read and Currently Reading; Already Read would complete the set. Actually, now that I think of it, this is probably where I would have said that a Read collection is vital: if you want to do anything with Read data, even something as simple as displaying it on the work pages, you can't assume that the absence of To Read and Currently Reading implies Read, as people often suggest.

maybe I'd like to see Tim recognize challenge groups and have him create a Challenge collection so there could be a Challenge module on the home page where folks could see books that have been read in challenges and challenge groups could be highlighted to drive traffic to those groups.

It's all interesting and it can all serve a purpose. But there has to be something other than "it's interesting."

I guess we can probably never get past this fundamental difference of opinion. I like the general idea of Challenge collections that are used to highlight the challenge groups on the homepage; people would have fun with that and get more enjoyment out of the site. What greater justification do you need?

why do we have pre-defined collections at all

It was based on a ton of user data - and years of requests - surveying. The years of it that Tim did with the users on this site - it was not random or based on what someone thought "might be interesting." In fact, at one point, "read" was on the table and was shot down over a lot of issues.

Zoe, enjoy your support. I tried to help. You clearly believe that the endless argument that you're willing to continue is the one way to get what you want and succeed. Don't bother to change tactics - it's only something you want.

It was based on a ton of user data - and years of requests - surveying. The years of it that Tim did with the users on this site

And there's been no further thought about it since before the feature went live. I think it's pretty standard to revisit this sort of thing after a certain period of time, to see how the initial decisions have worked out and whether any additions are needed.

Thank you for trying to help.

#39 "It was based on a ton of user data - and years of requests - surveying. The years of it that Tim did with the users on this site - it was not random or based on what someone thought "

Are you sure? Are you privy to LT staff discussions? From the general tone of the way Tim's generated new feature sin the past, I'd be a lot more confident of the assertion that Tim just put a few defaults in that seemed about right to him, listend for the loudest screams tweaked it once, and moved on. I very much doubt that any talk posts in 2006 or even 2007 were paid any attention to at all.

An 'I have read this' default collection that did something, just seems to me to be such a sensible addendum to all the other collections LT has. You are right that one can collect vast heaps of data, and not have any of it mean very much. But this is LT, it doesn't mean anythign to start with. But if it's interesting for some non-minority number of users, then manybe it's worth doing. Especially if it wouldn't take large amounts of coding time. For example the recommendation algorythms already exist. How hard is it to say instead of looking at "All collections" look only at 'I've read this'? We've got widgets for displaying covers on blogs. Why not only "I've read this" - and why not on the profile too!

41>

Well, prior to Collections, among the most common requests were for a "wishlist" and for "someplace for books I got from the library / borrowed from a friend / got rid of". That is, Wishlist and Read but Not Owned, so those were obvious default collections to include.

>42 I actually don't think "Read but Not Owned" was the best solution for that. If you want to distinguish the books you've read from the books you own, it makes more sense to have an Owned collection (e.g., Your Library) and a separate Read collection. The main thing is to have the unowned books out of the Library collection.

In my admittedly-small survey, Read but Not Owned was the least-used of the default collections.

43>

The main thing is to have the unowned books out of the Library collection.

True. And eons ago they were toying with a very simple model of two flags, a "read" and an "owned" status for each book that would generate four collections (with Unread + Owned ~ "To Read", Unread + Unowned ~ "Wishlist", and Read + Unowned ~ " Read but Not Owned"), which would have accomplished what most people were asking for much more simply than collections (and it would have given you your Read collection, but that ship has sailed).

>44 Yeah, I think I would probably have been happier with that earlier model. But I still don't think it's too late to add more basic default collections that could be useful for future features. Even though we have Read but Not Owned, it's not going to help with things like marking list completion, and it's not considered interesting enough to display on the Work pages (though apparently nothing is, these days--but I think that's just temporary). Read would be much more flexible.

>43 - In my admittedly-small survey, Read but Not Owned was the least-used of the default collections.

Reviewing this thread - your and stephmo's sampling is interesting but I should point out that there is a systemic error you are not able to correct for: the fact that one can "use" a collection without having it be active.

So, for instance, if either one of you looked at my profile as one of your samples - you might draw some false conclusions - i.e. you might think that I am not using the "Read but Not Owned" default collection. I am - it has 36 books in it - but I don't have it "active" so it doesn't display on my profile.

So, I think if you really wanted to know how people used different default/custom collections - you would probably have to ask them.

My "true" stats:

Active collections = 9 (2 default/7 custom)
*you only see 8 on my profile as one of the custom one's is empty when I am not actively working on a Combiners! project

Inactive collections = 10 (4 default/6 custom)
*all of these do have books assigned to them - and I am "using" them to some degree or another. However, one of the customs is a "Test" collection for figuring out particularly snarled combining problems and the books it it aren't "real".

Oh, good point. I wonder whether the error introduced by asking people directly would be less significant: in that case, we'd be restricted to the people who were willing to answer a question about Collections.

To go back to the original point of the thread, about deleting Member Recommendations:

A Member Recommendation recently appeared on my list that's pretty clearly not "in good faith". It's not malicious, per se, but it's proselytizing. Someone has recommended the Bible for The Hobbit. Unnecessary, not useful, and pretty clearly not in the spirit of MR. But there's nothing we can do about it.

Yeah, I noticed that too. But since there's no way to ignore member recommendations individually, I've pretty much stopped looking at them altogether. If The Powers That Be at least gives us the option to hide them (the Bible-for-Hobbit ones as well as the Harry-Potter-for-Harry-Potter redundancies), I might go back.

I'm new here and though I've read the information available about Recommendations (member and otherwise) I'm still not sure I understand. I really have no interest in this feature. Can it simply be ignored or is it something that needs 'tending' in some way?

Today was the first day I've had time to actually study this feature. It seems I have accumulated over 700 member recommendations -- clicking on No Thanks 700 times is a major waste of my time. Does the computer generate 1000 regardless of how often I say no thanks? I clicked it down to 900 today but it just felt silly.

Also, what is the process for recommending a book to another member? I have no intention of doing that -- except in a personal comment to that person or in a comment to a group -- and I never want to do it accidentally.

50>

Welcome to LibraryThing! You can't actually "No Thanks" a member recommendation, which is what we're complaining about here -- are you sure you weren't looking at the automatic recommendations? That said, you can certainly ignore either or both sections. I find automatic recommendations useful, sometimes, but have yet to find any signal in the noise of member recommendations.

(As for making one accidentally, just don't accidentally click on the "contribute your own recommendation" link at the bottom of the Member Recommendations for a work, accidentally type in a title, accidentally hit search, accidentally click on the title you don't want to accidentally recommend, and then accidentally click on "Make Recommendation". That is, you don't have to wory about accidental recommendations!)

Message edited by its author, Oct 27, 2009, 10:06pm.

48: I saw that Bible recommendation as well. FWIW, it was made by a 13-year old kid. The kid also rated the Hobbit only one star, and tagged it "horrible, too big, nasty." So maybe it was genuine, if not particularly well-conceived: those who like evil fantasy Hobbit need to see the light...? One wonders what he'd make of the Tolkien-Lewis friendship.

lorax: I'm accident-prone, so I worry about these things.
Good to know I'm safe here. Thank you for helping me.

52>

I'm pretty sure that "those who like evil nasty $book need to see the light" is specifically on the list of "not in good faith" recommendations.

The oddest part is that it's for the NIV. If you absolutely have to recommend a Bible for The Hobbit, the most appropriate choice would be the Jerusalem Bible, which contains Tolkien's translation of the book of Jonah.

Nasty Hobbitses.

Message edited by its author, Oct 28, 2009, 1:14pm.