metadata implicit and explicit in Tumblr

I wrote previously about Tumblr as a sharing-driven site, primarily used for photos.

Now that my Tumblr has been up for a while (it holds images I have licensed in the Creative Commons), I have more information about how Tumblr is used to share and tag photos.

You can think of Tumblr as a giant exercise in the crowdsourced assembly of photo collections.  People extract features that interest them, grouping photos in their own Tumblr streams.

So Tumblr has both explicit metadata, implicit metadata, official & unofficial collections, and social mechanisms for description and discovery.

For explicit metadata, Tumblr allows tags, creating a folksonomy of tags about an image (when you reblog an image, you can add your own tags).  You can also attach descriptive text to an image.

You can see both per-Tumblr tag views, e.g. http://rakerman.tumblr.com/tagged/books
as well as Tumblr-global tag views http://www.tumblr.com/tagged/books.  (Be aware that Tumblr has basically no concept of separating adult content, so take care when browsing even the most innocuous of tags.)  Search is entirely tag-based, e.g. when you enter "books" into the search box, it will return the tagged/books page.  It doesn't search the text in descriptions.  UPDATE 2011-06-23: Upon further examination it is not entirely clear how the search works.  Sometimes it returns hits that appear to be on description text, sometimes it returns hits for photos that don't appear to have any metadata at all.  ENDUPDATE

(Google does index the descriptive text, so you could search e.g. site:rakerman.tumblr.com reading.)

Tumblr also has (I presume manually-assembled) official collections for topics, e.g. http://www.tumblr.com/spotlight/books however these collections are not surfaced automatically in tag-based search (that is, if you search on books, it won't highlight the fact that there is also a books spotlight topic).

Where there is an additional layer of implict metadata is through the collections that people assemble themselves (to some extent using Tumblr to reblog photos in this way is one of the main use cases for the site).  For example, when I posted a picture of the Uppsala University Library it was discovered and reblogged by aveclivres, a Tumblr devoted to images with books in them.  (It was discovered even though I had no followers at the time, presumably by watching/searching the books tag.)

This particular image also gives a chance to see the "transclusion" reblogging effect, as a photo spreads in the Tumblrsphere.  It is last-in first-out, so it is essentially a most-recent-first activity stream specific to this photo (to this blog post, technically).

So in addition to the explicit metadata, we have the "collections" (aveclivres, bibliofila) that the post is in, which adds additional implicit metadata, "this is like these other objects in this collection".

"Reblogging" is like Twitter retweeting, and "liked"/favorites are similar to Twitter faves.  The reblogging mechanism is powerful as it lets the content spread, with popular images going viral in the same way that a popular tweet can go viral.

To some extent, this reblogging of images is a continuously running categorization "game" - more unstructured but also requiring less effort to create than a special online museum game site (see http://museumgames.pbworks.com/ for more information about this approach to online engagement).

Images present a special problem for categorization as unlike full-text it's much harder to use their contents for self-description except in a crude way ("mostly red", "lots of circles") - their "aboutness" requires human interpretation - interpretation which machines still struggle with, to the extent that image search mainly replies on surrounding context, user-supplied metadata, and image similarity.

The idea of posting images online for users to categorise is of course not new, it has happened through the Flickr Commons and other channels.  But Tumblr is a new way to tap into this user interest.

Tumblr has the concept of "following", which is also similar to Twitter - you see the stream of posts from people you follow on your dashboard, and it is one click from there to reblog or favourite a post you like.  In this way you can watch feeds of interest, giving a social channel for discovery of content.  In other ways Tumblr by default is not social in the way we think of traditional blogs - it doesn't have comments by default, and it has very limited mechanisms for feedback (the "ask me a question" option which allows for a question with a single response).

While Tumblr does provide powerful "Twitter for photography" features, its important to recognize that it is not a fully-featured photo management platform.  Here are some feature comparisons between Flickr and Tumblr.

Flickr has a multi-level access rights scheme per-photo (public, private, friends, family).  Tumblr is mainly for public blogging.

Flickr has a concept of adult ("unsafe") photos and "safe" public photos, with options to flag images.  Tumblr does not.  Tumblr has no advanced searching that will let you search only "safe" images.

Flickr records and exposes rich photo metadata, from date taken and location (if available) to the EXIF details about what camera was used, the exposure settings etc.  Tumblr does not surface any of these details, not even the date the image was taken.

Flickr has detailed per-photo rights information, including standard copyright and Creative Commons notices.  Tumblr does not.  You could apply a license to your entire Tumblr feed, but there are no built-in mechanisms to do so easily.

So be aware that in posting images to Tumblr you're moving from a personal photo collection environment where you have a lot of control about the individual photos and there is rich metadata, to basically a pure image sharing environment in Tumblr, where most if not all metadata is lost from the photo.

Tumblr is also about more than photos, it's a full microblogging environment where you can post text and video, but I've only looked at the photo aspect because its a common use of Tumblr.  There are lots of hybrid approaches as well, for example the US State Department Tumblr always features a prominent photo at the top, but has explanatory text as well.

UPDATE: Another great example of how Tumblr can provide infrastructure and a collection idea can go viral is http://dearphotograph.com/ -  the idea of overlaying old photos on top of current scenes.  A June 20, 2011 Globe and Mail article discusses how the idea started and the sudden spike of popularity for the site.  ENDUPDATE

UPDATE 2011-06-22: http://www.tumblr.com/explore is another way to discover top tags and content.  ENDUPDATE

UPDATE 2011-06-28: Tumblr has added the option to display basic photo EXIF metadata (and presumably could support richer EXIF metadata).  ENDUPDATE

Previously:
November 8, 2008  billions of photos