The chess board is the world; the pieces are the phenomena of the universe; the rules of the game are what we call the laws of Nature.
- T. H. Huxley

A great part of the authors’ experience has been web design, which involves organizing information into usable ‘categories’ for the client. As web sites become more social places to meet and share, as opposed to places to go seeking, having users extend the range of information architecture themselves, is both compelling and a little intimidating.

Tagging as an interface innovation (abstractdynamics) in today’s web architecture relies on user input to assign meaning to a resource, typically a link, image, or document. It has been adopted by a new breed of ‘social software systems, like delicious, flickr or Steve. Users tag resources with meaningful key words, and over time build individual tag sets that can be shared with other users, and with the system itself. Known as ‘folksonomy’ (Vander Wal) or ‘tagsonomy’, this trend leverages social intervention to give rise to a more meaningful web experience. (Shirky)

Meanwhile, efforts continue to pursue the construction of an underlying ontological framework forming a stable basis of meaning for all web users. It is felt that an all encompassing ontology users can rely on to navigate, and that can be interpreted by machines will be the most effective environment to make sense of this expanding space.

The authors see a need for a combination of machine readable structure, which can coexist with socially driven input. Folksonomies greatest strength and at the same time its greatest weakness, is the variation of terms that users can assign or tag to a resource. Semantic differences such as using plural tenses are responsible for much of the disparity, and cause the separation of ideas that should otherwise be grouped together. Also, there is the capability to dilute the meaning of a tag through inconsistent or malicious use, such as spam. Ontology and other hierarchical based approaches are very rigid and do not make possible responsive information spaces that reflects the natural world we exist within.

Cellular Automata Systems (CAS) such as Conway’s Game of Life or flocking simulations seemed good departure point for considering other social arrangements that have inherent structure. A CAS is a dynamic system that arranges itself based on its current state, and the state of the elements within the system. Each element follows a simple set of rules that define how they communicate with each other, and how they respond when they encounter another element. By following the rules an order can emerge within the system that is not inherent within the rules themselves.

Both CAS and folksonomy are self-organizing and rely on the isolated acts of the individual to create a higher-level order. However, the organizational structure in many of the popular implementations is very low level when compared to other bottom up systems like CA. The difference being CAS is governed by a set of universal rules, while the average user and tagging interface are not. It is the rules that make (CAS) work.

This paper discusses potential opportunities to improve the structure and meaning of a resources tag set by adding easy to use rules to the process of tagging. We examine existing rules in popular folksonomy websites and outline some of the interface patterns that have emerged. The authors discuss ‘adding metadata to the metadata’ by the weighting of tags or sequencing of tags. This study will frame the guidelines of a series of workshops being designed to gather data that can inform further exploration of this topic.

The rules as we know them

Tagsonomies are regarded as intuitive ways to collaboratively classify resources. Most of the literature over the past year however has targeted the weaknesses in using them, and the disparity among conventions users must overcome to successfully input their metadata (37signals, Return of design, Hey Norton). Our research considers these conventions the primary rules that must be learned in order to participate effectively. Some of these rules are driven by the interface and overall usability of the site, others originate from the social networks and cultural practices of the participants. In considering existing rules, we wanted an outline of common UI patterns as an initial departure point.

Looking at the most common type of site where many people can tag a single digital artifact, known as broad folksonomies (Vander Wal) : 43things; Amazon; blinklist; CiteUlike; del.icio.us; Furl; last.fm; technorati; upcoming . Other types fall into some loose categories. Systems that one person tags many digital artifacts, narrow folksonomies: Flickr (now can be customized); Gmail, and systems that the responsibility is shared: consummating.

Adding Tags: spaces and quotes vs. commas and hyphens
The first step in creating folksonomies is getting users to add terms they find meaningful to resources. The interface element most commonly used for adding tags is a single line text input field.

del.icio.us

Users can freely add keywords here but must enter the information using the correct format. One of the challenges for users is remembering how to separate the tags being applied. In some cases commas are used, in others, spaces. This presents the problem of having multi- word tags, such as wordswithnospaces. Other systems, have tried to overcome this by incorporating quotes to contain “multiple word tags”. It was found in research done on STEVE that the size of this input field had a direct affect on how users would enter information. They found that allowing users a larger input area than the standard one line for instance, would result in longer narratives being used to describe the objects as opposed to the list of distinct keywords they were looking for.

Tag Tools: pop up vs. browser
The place from which tags are entered is changing as well. Early on, pop up windows where a common way for users to tag web pages, in social bookmarking tools like flickr and delicious. Increasingly, these interactions are being supported by the browser itself. (Performancing, Flock)

performancing

This shift signifies the growing importance and ubiquity of tagging removing yet another barrier. Flickr affords two unique tag input tools, one as a sidebar link that can be expanded to add links one at a time. Also, they provide a downloadable application that supports tagging at the upload stage, and handles other server intensive tasks client side such as resizing.

Ambient Interface: tag a little vs. tag a lot
While user attention is typically focused on the primary tag field, where tags can be freely added, there is a growing trend to afford users other means of adding metadata.

last.fm

As a users’ labeling behavior becomes apparent, common tags can be presented underneath or along side the labeling field that can be clicked to add them to the current item. Also, popular tags from other users can be applied (in a broad folksonomy) in a similar manner. ‘Smart’ text popup displays are also appearing with the first letter you type trying to guess the tag you will use before you finish typing and giving users the option to click instead of keyboard.

Tag context: tagging music vs. books vs. ourselves

consumating

The community of practice interacting with the resources in question brings an existing set of linguistic habits, a “loosely-joined vocabulary” or in our case, rules. The authors have observed some interesting trends in tagging across different systems and hypothesize that habits will alter greatly depending on the context of the thing being tagged. This is the case at the dating site consummating where users are asked to tag themselves, and each other. Systems like this have different implications than those where the resources are ones private email or public photo collection. As folksonomy is applied in more diverse and unique situations, more variety and flavors of tag vocabularies will be specialized to suit the domain.

Electronic communities have evolved using communication tools such as email, discussion boards and chat spaces, and unique vocabularies have emerged in each that allow users efficient ‘keyboard shortcuts’ to convey significant concepts. “lol”, “brb”, “”, “foo” Although similar trends have not yet been observed in existing tag practice, (refs!) we suspect that in general tags will develop some unique syntactic rules. Particularly pertaining to specific cultural domains.

Tag feedback: clouds vs. lists
Tag sets are displayed in a variety of ways to users. A recent trend has been to use tagclouds where more common tags appear in a larger font size.

tagcloud

Alphabetically sorted lists are still popular and widely used. How tags are displayed has potential to support overall findability, but noting our own experience it seems that tags in themselves do not necessarily make items more findable. You still have to remember what tags you used originally if you are set upon finding a very specific item. However tags as loosely defined terms for browsing activities seem much more successful.

Considering New Rules

We are considering a number of additional rules that could be implemented into a system to modify the relationships between the resource and its tags. Before considering a new rule we wanted to be sure that it would fit the purpose of our research by improving the findability of a resource. The rules that we are considering fall into two different strategies: strengthening relations and grouping/clustering.

Strengthening relations is a strategy of applying values to the edges that connect the resource and the tag. These values would act as weights, where stronger connections would have a higher value and weaker connections would be lower. Modifying the relation between a tag and a resource should allow the system to return information in terms of relevance, as well as popularity and chronologically.

Grouping/clustering looks at ways to organize like resources together into groups based on their similarity. These groups form new structures in the data that are not implicitly built into the ontology. These groups could be offered to the user as options for extending or refining their search, or as a method to further explore the tag space.

Weighting
The strength of the relationship between a resource and a tag can be determined a number of ways, both from within the system and through external intervention.

A resource could keep track of a tags frequency, or how many times it has been applied. Terms that are used more often could be considered more important, or given a higher value by the resource, aligning it closer to the tag. This system is similar to the popularity rank already in place at delicious, but would extend the system to take into account the usage of a tag on an individual resource by all users.

Another option is to allow the user to define the relationship value between a resource and its tag. As the user applies a tag to a resource they would be given the opportunity to apply a value to that tag. Adding to the users workload is not without its potential pitfalls. We need to be careful not to introduce a form “post activation analysis paralysis”, (Sinha) where the user is afraid they will make the wrong choice.

Clustering
Like relation values, clusters formation can be automated by the system, or created manually by the user.

A tagging system could use a hierarchical clustering algorithm (Jain) to create groups based on the similarity of each resources tag set. A hierarchical clustering algorithm determines the similarities between two resources by taking the number of correlated terms and comparing them to the number of non-related terms. The more terms the resources share, the more they have in common,

Relationships could also be based on language. For example, a system could create a strong correlation between two resources that use the singular and plural of a term. This would allow a system to treat map and maps, or category and categories as the same term, and return results for both. A more ambitious approach could use more elaborate language structures such as synonyms to create clusters.

Mapping user-tagging patterns could also be used to form groups. It has been observed (Golder, Huberman) that users tend to use tags in a particular order, favoring general tags initially, and then getting more specific with each subsequent tag creating a parent child relationship between each tag. A tagging system could create clusters based on shared genealogies.

The process of group creation could also be offset onto the user, allowing t he user to define the relationships between tags by placing them in the same group. Delicious has implemented this strategy with bundles. Selecting a bundle will return all of the resources that have been tagged by a member of that bundle.

Conclusions

This paper reviewed potential opportunities for improving the structure and meaning of a resources tag set by adding a layer of rules to the process of tagging.

The properties of cellular automata have inspired our vision of a dynamic tag space that can reveal patterns of behavior as it evolves. Self-organizing systems rely on the isolated actions of the individual to create a higher-level order. It is the interface and cultural rules that define those interactions in a tagging system and thereby can impact the structure of information they are applied to. Both interface conventions and cultural context drive the users’ motivation for tagging resources in an electronic space. Exploring existing designs that allow users to easily enter tags, we have established a baseline set of rules as they exist today. Through brainstorming, we have identified some key areas of the user experience for consideration and usability testing in our participatory design workshops. They will inform further exploration of this topic and open up new avenues of interface design for consideration in future systems. The continued success of tagsonomies will be measured by the return on investment that user gain for their efforts.

There is a website of the ongoing workshop and results http://te.chni.ca/tagEverything/

The authors blog related links and thoughts http://te.chni.ca/waypoint/

Some systems that are applying tagging in interesting and innovative ways: Graphalicious; extisp.ico.us; tag.alicio.us; Googlebase; STEVE; The ESP game; Wanabo

References

A shortlist for this paper. Full project credits continue to get posted online

The Structure of Collaborative Tagging Systems
Golder S, Huberman BA 2005

Rashmi Sinha’s Blog (http://www.rashmisinha.com/)

Jain, A.K., M.N. Murty, P.J. Flynn. “Data Clustering: A Review”. ACM Computing Surveys. Vol 31, No.3. September1999

Abstract Dynamics Blog, William Blaze; August 29, 2005 (http://www.abstractdynamics.org/archives/2005/08/29/tags_bottom_up.html)

Thomas Vander Wal Blog, January 18, 2005
(http://www.vanderwal.net/random/entrysel.php?blog=1622)

You’re it! Blog, Clay Shirky; June 5th, 2005
(http://tagsonomy.com/index.php/tagging-saves-categorization-from-itself/)