Big vs. Bold in Tag Mirror — What's the Difference?

My intuitive thoughts when looking at the tag mirror are that bigger words are more common/significant, but I would also think that words in bold would be more common/significant. Sometimes bigger words are in bold, but not always.

For example, 'nonverbal communication' and 'early childhood education', in my tag mirror, are some of the smallest terms, but are in bold. 'William Shakespeare' and 'horror' are both very large terms, but neither is in bold.

I think bigger means more common in my library, but am I correct in that? And what exactly does bold mean?

i find it quite amazing that no one in LT support has answered this question. I have tried to get an answer to it for several years, including today. All they could come up with was the original discussion by Tim talking about the Tag Mirror, and links to Talk which is a compilation of questions similar about Tag Mirror.

I had to laugh as that link came up with @JoeB1934 asking, in essence, this question over the years. I have just moved on and clicked each tag to get the numbers.

It is also interesting that several support people have defended that there isn't a toggle to display the numbers. They say, oh that would defeat the intent of this font display concept. OK, how about explaining what the fonts are intended to say, and tell us about other members that actually use the font concept.

>2 JoeB1934: Hmm, maybe it has something to do with the prevalence index you see when looking on your books with a certain tag?

Or it might indicate that you are also using the tag on the same books as others?

Or maybe it is meant to be an eternal mystery?

Neither bold nor big represents 'you also use the same tag as others' because as of writing this post, I did not have any tags I had put on my books which showed up in the tag mirror at all, and I had large tags and bold ones.

>4 GraceCollection: That is the normal behavior, in that the tag names are what other readers place on your books, even if you don't have any tags in your whole library. When I rebuild my library, I start with all my books without any tags on anything.

LT knows what books I have and is telling me what tags on those books have been used by other readers. I have a list of tags that I want to know if they are on these books. When you click on the tag name in the mirror up pops the list of which specific books in my library have that tag name attached to them by other members.

I use the mirror every day to find tags for my books and I assign that tag to my books through a complicated export from LT and import to a spreadsheet. After processing this import into my books, I import those tags back into my library which doesn't have any tags.

Tim is strongly opposed to what I do, as he wants every member to choose their own tags. As far as I know this FONT concept is his attempt to tell me something about that tag by way of the prevalence numbers. They do say how many readers of that book used that tag.

I absolutely do not see what can be wrong about my assigning a tag to my books if other members have done so. With a library of 1900 books does he expect me to do the mental gymnastics to do that any other way. I don't use a tag for a book if it isn't one of my 30+ tags that I want to know if other members use one of those tags.

If I create a collection of the books, I read this month I can apply the mirror and find out ALL of the tags other readers placed on the book. If I read a newly published book there often aren't any tags there.

A real issue with using my method is that the number of readers who place a specific tag on a book is extremely low when compared to the number of readers of the book. Take a tag such as 'historical mystery' and the number of readers who tagged the book might easily be a small percentage of the readers of the book.

Speaking of imports, if anyone tries to do what I do, pay special attention to the sample.csv file for doing imports. I have been doing these imports for years and suddenly they didn't work anymore.

After an extended squabble with LT it finally developed that they are working on improving the import process and the method of ENFORCING the sample structure changed without any notice that it was different now.

I noticed in a tag cloud on a specific book, where each tag save one was only used once, they had different sizes. I assume the rules for tag mirror and the clouds on the work/author page are the same.

I feel even more baffled!



The work is Queen Hildegarde, but I've added the book with my own tags recently, so it may change again at some point, depending on when the tag cloud index runs.

>6 keristars: If you look at the book page there are only 22 members with this book and 2 reviews. This is the problem I mentioned earlier, that not many readers actually go to the trouble to place tags on the books. Probably only for those that have reviewed the book.

Ignore the font info and just look at the tag names, that is the most you can get out of this mirror analyses that I have found to be useful. In this case I have found it useful to go to Goodreads, where there are usually far more readers. They produce a set of tags which are very useful to me when LT doesn't have sufficient readers.

GR has 76 ratings and 18 reviews. The only tag they have is 19th Century.

I import ALL of my library into GR just to obtain clarification on books where LT has insufficient data.

GR produces the numbers for every tag and all tags.

I actually prefer the LT mirror over the GR listing, which is mostly useful for situations where LT doesn't have enough data.

Using the mirror has led me to inclusion of tags which originally, I didn't think I was interested in. Specifically, tags 'Family, Relationships, Romance' come to mind.
The mirror also led me to using 'Literature', and 'Historical Fiction' as being critical to my reading.

The bottom line is that the mirror revealed to me why I preferred certain books and how to find more of them.

While I can't tell you the exact algorithm used to compute the sizes for the tagclouds (it's precomputed on a schedule by a worker-process that isn't one of mine), I can tell you that the top ¼ of the tags within a cloud get bolded, just to increase their visual significance.

>8 conceptDawg: Thanks for that update. Why isn't there a link to a discussion about this Bolding process. A lot of people would have their questions answered.

I don't know what all the rules are, but it must not be strictly numbers. It appears that prevalence is a big factor.

I was astounded when a tag that is appropriate for very few of my books is very bold. That tag is 'Paleo Indians'. This is for 53 books out of 1950 books. The prevalence values are over 10%, which means to me that these books really are about Paleo Indians.

At times I have been tempted to make a set of my personal favorite tags just for all of the highlighted tags. However, most of them already are my favorites. Some of my favorites came to me after I saw them highlighted.

I would really appreciate a thread amongst the members, especially who use and enjoy the highlighting features. How do they use this somewhat mysterious feature?

Okay, so the various tag clouds work in different ways. This is mostly about the tag mirror.

First, ALL tag clouds are attractive representations of the data. They are not a 1:1 copy of the data. They are designed to make you understand something, and to enjoy it, not to substitute for a spreadsheet. If you want the full information, click the options to see all the tags and their counts.

You can't do this for the tag mirror, because it's uniquely impressionistic. Why? Speed, memory—and not being boring.

The base for the tag mirror is taking all your works and looking at the tags applied to them. But, for speed and memory reasons, we only take the top 20 tags on each work, excluding singletons.(1) The downside of this is that we are giving you an impression, not an exact count. For that reason, and because of the significance question, we don't print out the numbers on the tag mirror.

Once you have the counts, you need to turn them into pixel sizes. But whatever the cloud, you need all kinds of special rules to deal with the wide variety of numbers and their distributions. You don't want a work with ten tags, but one is 20x as popular, to have 9 tags at 7px, and one at 140px. So it sets some upper and lower limits, and flattens the numbers mathematically. The result is intended to showcase the relative prevalence of the tag, but the exact pixel height does not correlate exactly with the number of uses. It's dozens of lines of code.

As for boldness, for the tag mirror, but not for everything, the answer is that the top 10% by "significance" are bolded. Significance is defined for a tag as:

Count of times on all your works
divided by
SQRT( count of times across all works on LibraryThing + 1) + 10

The idea here is to identify the tags that are particularly common for your works, but not necessarily common overall. So, for example, if you have a lot of books about Colonial America, the "history" tag will undoubtedly be far, far higher than the "Colonial America" tag. But the latter is more pertinent or significant.

This is always the tension in tags—and indeed in recommendation algorithms too! If everything were based on the raw count, then the best recommendation for every book would be Harry Potter. If we used a significance metric (how common on these books as against overall), then all recommendations would be super-obscure stuff. So we come up with compromises of various sorts.

For the tag mirror—but not other tag clouds—we fiddle with the significance by a hand factor, which is whether or not the tag has been staff approved for use in libraries. This approval process approves tags about subjects, topics, moods, genres, authors, series—anything broadly meaningful, and rejects tags that are purely personal "read in 2023," "import," "goodreads" or edition-based tags "paperback," "kindle," because the tag cloud is about works, not editions. If the tag is approved, it's a 20% benefit to its significance. If it's rejected it's a 50% penalty. This prevents tag mirrors from having a lot of personal stuff—either yours or someone else's who shares a lot of books. So, for example, if you tag every one of your books "at mom's house," the rareness of that tag combined with it being on every book you own might make it a top, bolded tag-mirror tag, and we don't want that so much.

---

1. For a sense of scale, the work 1984 has 9,621 unique tags applied to it—unique AFTER tag combination. To Kill a Mockingbird has 9,429. And together they have 16,420—a lot less overlap than you'd want! You see the problem? If we didn't take the top X tags for each work, we'd be doing math on millions of unique tags to generate a single tag mirror for anyone with a normal library of common books.

>10 timspalding: Thank you for this explanation, it's marvellous!

>5 JoeB1934: Hi, I do know what Tag Mirror does, generally. My comment in >4 GraceCollection: was about >3 bnielsen:'s theory that either size of test or boldness/lack thereof 'might indicate that you are also using the tag on the same books as others'.

As far as I know this FONT concept is his attempt to tell me something about that tag by way of the prevalence numbers. They do say how many readers of that book used that tag. As I said in >1 GraceCollection:, I posted this to try to discover what the difference is between bold text and largely sized text. There is no way both of these features can represent 'how many readers of that book used that tag' because, as I said in >1 GraceCollection:, some tags are either big or bold, and not both. If both big and bold meant the same thing, you couldn't have any tags that were one but not the other.

>10 timspalding: This is very helpful and exactly answers my question! Thank you for taking the time to explain.

>10 timspalding: Thank you for this explanation, now I can proceed with how I will use the Tag Mirror. It, along with Tagmash, have been the key elements in my understanding of my book reading.

I can also tell you that the Goodreads system for tagging, which is strictly numbers based doesn't produce the quality of knowledge about tags that LT does.

I have noticed many times that I preferred the LT tags analysis over the GR tags. I didn't know if their readers are different than theirs, or what.

This approach you have, in my experience does uncover tags that their number-based system doesn't find.

GR and LT do match fairly well with major genres, but not at all in what I call sub-genres.

FWIW, I love (love!) data, so I'm minded to come up with other measures, if you have ideas for what you want me to show. The big problem is working with the constraints of speed and memory. Programming is odd—something you think it easy may be hard, something hard may be easy—so feel free to throw ideas out there and I'll think about them.

>15 timspalding: I too, am really interested in data analysis of all types. I don't have anything like the depth that you obviously possess. I have used analysts like yourself to explore processes that help in a project I might have. I have two questions about this tag analysis that are relevant to myself.

First, I have built a list of about 40 tags that are of special interest to myself. When I do a Tag Mirror, I search for those specific tags first thing. Would it be possible for LT to have a personalized list of those tags and just do the analysis for them.

Currently, I use the ability to focus on a collection to limit the mirror scan and I use that feature a lot. This would be a similar way to reduce the computer resources.

Second, it always astounds me how few readers actually bother placing tags, other than what I call bookkeeping, or personalized tags, which is very common. In GR these are bookshelves, similar to your idea of library definitions.

What if my personalized list of tags were thought of as the eligible list of library shelves.

This is like a hand mirror for your face versus a full body mirror.

>15 timspalding: If the information in >10 timspalding: was included somewhere as an explanation on the tag mirror page, I think it would go a long way to clearing up confusion! As to what else I'd like to see:
- "Missing" tags (probably not the best title): Tags that are normally present in similar libraries to yours, which you don't have or have very few of. For example, "Most people whose significant tags include education also have a large amount of books tagged music, but you only have two"
- Similarly, tags that have an average prevalence of some certain percentage, that are lacking or maybe even totally absent in your collection. This would be especially helpful to community libraries, I think, to find spots that may have accidentally been overlooked, but nonetheless would be interesting to anyone who finds data as interesting as we do! Example: "Most people have an average of 15% romance books, but only .5% of your library has that tag!"
- Unique tag combinations: Tags that are both significant in your library, but are pretty rarely both significant in a library. "Most users whose significant tags include picture books don't also have horror as a significant tag, but you have both!"
- Top library: If you have the most books (either per number only or per percentage of your library) in a tag, that would be really interesting information to know. For example, "You are in the top 10% of users for amount of your library with the weird fiction tag!"

Some data I'd love to see that is slightly less related to tag mirror:
- unique tag combination book recommendations: Out of significant tags in the user's library, those who have very few tagmash results would make an incredibly interesting recommendation list. For example, "You have both Roman architecture and religious psychology in your significant tags. Here are the only 3 books on LT which are tagged with both Roman architecture and religious psychology."
- Tag search in your books: When I search for a tag on the general LT search, I can scroll through the list of books sorted by most frequently tagged and try to locate ones that have a symbol indicating I own a copy, but I'd much prefer to be able to hit a switch or do a search on some other part of the site so that the results I get are only among my books. I can search my books in my library, of course, but not using frequent tags; only using my own info.

What is a tag mirror versus the tag cloud?

>18 TheLibraryAnn: The tag cloud shows how you’ve tagged your books. The tag mirror shows how other members have tagged your books.