Tag Camp 2005

Uzanto

Digital categorization

Stage 1Multiple

concepts are activated

Stage 2Choose ONE of

the activated concepts.

Categorize it!

Note the chosen concept

Stage 0Object worth remembering

(article, image…)

Cognitive process behind categorization

Analysis-Paralysis!

Uzanto

Tagging is simpler

Stage 1Multiple

concepts are activated

Tag i t!Write down all

activated concepts

Stage 0Object worth remembering

(article, image…)

Cognitive process behind tagging

Uzanto

Tagging: Input versus Output

Output Structure

Input EffortLow High

Low

High

Tag Clouds

Hierarchicalcategorization

Facets

Clusters

Collaborative Filtering

If we could get to Facets / clusters from tags, it would be best of both worlds

Uzanto

Getting at structure through tags: (a) Card Sorting

http://www.rashmisinha.com/archives/05_02/tag-sorting.html

Uzanto

Getting at structure: (b) Collaborative Filtering

• Challenges with Input: Getting lazy users to provide input (some indication of interest) Implicit (infer user intention from behavior). e.g., past

purchasing behavior on Amazon Explicit (powerful, difficult to get people to rate stuff) e.g.,

ratings on Pandora, MediaUnbound)

• Challenges with interface Hard to convey what system is doing (no direct cause &

effect as with search) Need to make system logic transparent (e.g., Amazon)

• Users who liked X, also liked Y• Users who searched for X, also liked Y• Users who bought X, also bought Y

Uzanto

Varieties of Collaborative Filtering with Tags• Tags as an indicator of interest

Tagged url as input, url as output

People who like this url also like these other urls Tagged

Uzanto

Using tags as semantic information (with CF algorithms)• Tags as input: Tags as output

(People who use similar tags also use)

• Tagged urls as input: Tagged urls as output People who tagged similar urls’ also tagged these urls

• Tag-url conjunction as input: Tagged url’s as output People who tagged similar url’s with similar tags, also

tagged other url’s with same tags

Uzanto

User Experience in such Search/Browse interfaces

•More of a controlled experience

•Every movement (forward, making a turn, backwards) is a conscious choice. (need information at every step)

•User might make mistakes, and retract (go back) a step or two or start again. Each of these is a conscious choice.

Like being in control of a car

Uzanto

User Experience with Recommender Systems

•-user has less control over specifics of the interaction.

•System does not provide information about specifics of action

•more of the black box model (some input from user, output from systems).

Like riding a roller-coaster

Uzanto

Tags & Clusters

This idea is not new!

Uzanto

Clusters on Flickr

• Positives: No effort to name the clusters (makes clusters easy to understand) No fancy visual ways to show the clusters, simple text and image

based Clustering tags, rather than items (computationally easier, easier

to understand( No Other or Misc category

• Negatives Hard to understand the overall space and the individual

categories Much harder to do with anything but images

Uzanto

Tags & Facets (the holy grail?)

• You have several orthogonal classifications. For example, Wine.com

Wine can be described by type, region, price, vintage.

Instead of putting all this in one large hierarchy, you can have several hierarchies

Natural afffinity between tags and facets• With tags you don’t need to know where stuff lives,

just know a few of its descriptors• You can have as many facets as you want• There are multiple ways to access a piece of

content

Uzanto

From tags to Facets

• More than one way Compute facets from tags in some intelligent manner

RawSugar• Uses tags• Some sort of a taxonomy at the back end to make

sense of tags

http://www.rawsugar.com/collections/oferben

Uzanto

Chandler approach

• People tag mail and other items

• Can promote tags to facets