A cognitive analysis of tagging

A cognitive analysis of tagging

(or how the lower cognitive cost of tagging makes it popular)

At the start, let me confess that I struggled with this topic. From
my first encounter with tagging (on systems such as del.icio.us &
flickr), I could feel how easy it was to tag. But it took me a while to
understand the cognitive processes at work. What follows is Rashmi’s theory of tagging
– my hypothesis about the cognitive process that kicks into place when
we tag an item, and how this differs than the process of categorizing.
In doing so, my hope is to explain the increasing popularity of
tagging, and offer some ideas regarding the design of tagging /
categorization systems.

My ideas are mostly based on my observations about how people tag
and relating it to on academic research in cognitive psychology and
anthropology. This is a first version, which I expect to revise as I
learn more. Feedback is very welcome.

The rapid growth of tagging(1)in the last year is testament to how
easy and enjoyable people find the tagging process. The question is how
to explain it at the cognitive level. In search for a cognitive
explanation of tagging, I went back to my dusty cognitive psychology
textbooks. This is what I learnt.(2)

Categorization is a 2-stage process.

Stage 1: Related Category Activation The first
stage is the computation of similarity between the item and candidate
concepts. For example, I come across the book « Snowcrash » in my
library. Immediately a number of related semantic concepts get
activated: « book », « science fiction », « Neal Stephenson », « Zodiac ».
Other concepts might be more personal; e.g., « favorite author »,
« airplane trip ». Still other concepts activated might be more about the
physical characteristics, e.g., « paperback », « bad condition ».

How
do we know this? Cognitive psychologists have explored this phenomenon
by asking by asking people to list semantic associations with an
object, and mapping the type and frequency of associations. Another
method is to use implicit memory measures
to probe what concepts have been activated. With the advent of fMRI, it
is possible to correlate such concept activation to changes in blood
flow to difference parts of the brain. The details of this are not
relevant for the present discussion, what’s relevant is that there is
broad agreement about such conceptual activation in cognitive
psychology.

So far we have learnt how related concepts are activated. Writing
down some of these concepts is easy enough. With tagging there is no
filtering involved at this stage, you can note as many of those
associations as you want. This is how tagging works, cognitively
speaking. Yes, it’s that simple.

On the other hand, the work for categorization is just beginning.

Stage 2: The decisionNow that we have candidate
categories, we need to make the DECISION. What category is the right
one? Cognitively, the process is fairly simple – you compute similarity
between present item and candidate categories. A function called Shepard-Luce(ppt) describes how people make this decision.

(The process sounds intimidating, but generally its not. Choosing
the best category is something we do all the time. We see an animal –
it could be a dog or wolf. We make a quick judgment. This is a basic
cognitive process – putting things into categories. Even birds can do
it. When I was at graduate school, the Professor in the lab next door studied how pigeons make categorization decisions. There is evidence that babies can categorize.)

Cognitively, we are equipped to handle making category decisions.
So, why do we find this so difficult, especially in the digital realm –
to put our email into folders, categorize our bookmarks, sort our
documents. Here are some factors that lead to what I call the « post-activation analysis paralysis ».

First, there is less cultural consensus around
items we categorize in the digital domain. Categorization is often
based on cultural knowledge. For example, over the years we learn the
cultural consensus regarding the boundary between wolf and dog, couch
and chair, fruit and vegetable. With digital objects, there is less
cultural knowledge about the categories – in fact, one purpose that
tagging serves is transmitting cultural knowledge about our constantly
evolving digital lives. (This is an interesting topic in itself and
deserves a whole other essay).

In the digital world, we don’t just categorize an object, we also optimize its future findability. We need to consider not just the most likely category, but also where we are most likely to look for the item at the time of finding. These two questions might lead to conflicting answers, and complicate the categorization process.

Also, with digital objects, it’s not just adhoc categorization – put an object into a category, any category that comes to mind. We need to consider the overall categorical scheme.
Is my scheme becoming unbalanced? Do I have too many items in one
category, and too few in another? If I put everything in one category,
I will never be able to find anything. Do I need a new category for
this item? Does it even fit into this scheme?

The need to consider the overall categorical scheme is much more
important in the digital realm, than in everyday categorization
decisions. For example, I come across an animal; I categorize it as a
dog. I don’t need to worry that my mental dog category has become too
large, that I might need subcategories. Cognitively, it is sufficient
to make local decisions about objects we encounter,
the brain does the magic computation, and my animal taxonomy evolves.
As I become an expert on dogs I evolve sub-categories for spaniels,
dachshunds and terriers, without explicitly thinking about the
structure. Next time, I encounter a terrier, magically, the terrier
subcategory gets activated. Think of how much work it would take for
something like to happen with our email folders.

Finally, there are no second chances in categorizing digital objects.
Well there are – but those are fairly expensive. You need to go into
the first category, retrieve the item, and put it into the second. This
is where user interface for categorization comes into play – most
systems assume that you are done with an item once you categorize it.
It’s taken away from you. The brilliance of Gmail was to separate the
tagging from the archiving.

Start thinking of all this and you land into « post activation analysis paralysis ».
A state of fear that you will make the wrong decision. And the item
will be lost forever – it will land in some deep well, some hard to
access branch of the tree and disappear from your view and attention.

We come back to the question that we started with – why is tagging simpler. In my opinion, tagging eliminates the decision – (choosing the right category),
and takes away the analysis-paralysis stage for most people. (Note that
some people might still freeze up in deciding between different tags,
or figuring out ways to optimize future findability. These are valid
concerns that tagging systems can address better than they do now).

Another observation about tagging – it provides immediate self and social feedback.
Each tag tells you a little about what you are interested in. And you
find out the social context for that bit of self-knowledge. How do
others view that item? Together this piecemeal feedback creates a cycle
of positive reinforcement, so that you are motivated to tag even more.
This might not make tagging easier, but it does make it more fun.

To conclude, the beauty of tagging is that it taps into an existing cognitive process without adding add much cognitive cost.
At the cognitive level, people already make local, conceptual
observations. Tagging decouples these conceptual observations from
concerns about the overall categorical scheme. The challenge for
tagging systems is to then do what the brain does – intelligent
computation to make sense of these local observations, and an
efficient, predictable way to ensure findability.