HT06, Tagging Paper, Taxonomy

Today I’m presenting a paper I coauthored with Mor Naaman, Marc Davis and danah boyd entitled “HT06, Tagging Paper, Taxonomy, Flickr, Academic Article, ToRead.” It’s possibly the least memorable title in ACM history, but it seemed like a good idea at the time. This publication is a position paper, and as such is focused on

In the paper we present two taxonomies of tagging, the first dedicated to design decisions in tagging systems, and the second to the incentives that drive people to tag therein. We also present a short study of Flickr and compare to Golder’s del.icio.us analysis to show how some of our distinctions may affect the behavior of taggers within the system. Here are some of the important distinctions we found among tagging systems.

Tagging Rights

One of the most important characteristics of a tagging system are the permissions given to taggers. A tagging system can be restricted to self-tagging, where the users allowed to tag are those who created the media (e.g., Technorati or YouTube), free-for-all, where anyone can tag any object (e.g., del.icio.us), or permission-based, where the owner determines who can tag (e.g. Flickr).

Tagging Support

Another important distiction is the extent to which the system aids in the tagging process. In the base case, blind-tagging, the system provides no help whatsoever (e.g., Flickr and Upcoming). In the case of suggestive tagging, the system provides some suggestions for possible tags at the point of tag entry (e.g., del.icio.us or MyWeb 2.0), and viewable tagging systems show previous tags at entry time (e.g., Yahoo! Podcasts).

Aggregation

Another feature of group dynamics and possibly the largest differentiator of tag systems is in how the tags are grouped. In set-based tagging system, all of the tags go into one set, as with Flickr or Upcoming, whereas bag-type tagging systems, each tag entry is distinct, and tags are aggregated based on their frequency. This is the model used by del.icio.us.

Type of Object

One more important distinction is the type of object being tagged. These can be either textual (del.icio.us, MyWeb 2.0, Technorati, etc.), or non-textual (Flickr, LastFM, YouTube, etc.). In the case of the former, tags represent an alternative feature for text algorithms, while the latter they may represent the only textual metadata.

The paper also presents a few more distinctions, although they are less significant. In the case of user incentives, we have identified the following categories:

Future Retrieval

The most cited reason for tagging is to allow for the future, personal retrieval of objects. This is the selling point of many social bookmarking tools, along with a method for organizing blog posts and photos.

Contribution and Sharing

This refers to the process of creating groups or lists for sharing to friends or unknown audiences, as with groups in Flickr or lists in del.icio.us.

Attract Attention

The third incentive refers to tagging for the purpose of attracting attention or enabling search. In this case, the user is tagging with the hope of attracting users to one of their pieces of content. Because full-text search was not reliable early in Flickr, this was an important reason for people to tag, as it allowed other users to find their content.

Again, the paper provides more detail and a few more categories. You can also take a look at my slides for the presentation (1.3 MB).