By Gary Hsieh
Searching for information online has become an integral part of our everyday lives. However, sometimes we don’t know the specific search terms to use, while other times, the specific infor- mation we’re seeking hasn’t been recorded online yet.
What we often resort to, after a few minutes of searching, is asking
This type of information-seeking behavior is one of the primary reasons social question and answer (Q&A) sites have become more popular. In fact, these sites—Yahoo! Answers is one example—enable tens of
thousands of user-asked questions to be answered daily by other users.
The basic premise of these sites is that anyone who has a question
can post it, and others in the community can respond and share their
expertise or knowledge.
However, if you’ve ever used one of these social Q&A services,
you’ll know that they are not designed to accommodate differences in
individuals’ needs and constraints, which can result in inefficiency.
People with questions who are in dire need of answers typically have
no way to indicate their urgency and may not get an answers in time,
while people who answer questions may feel overwhelmed by all the
questions, especially if they’re directed by potentially disruptive communication channels, such as instant messaging.
Quickly browse through Yahoo! Answers and you’ll notice that
seemingly important questions are presented alongside a substantial
number of frivolous non-questions. For example, right below one sincere question, “Where can I go in Palm Bay, Florida, to get assistance
with deposit money for an apartment?” there is a more frivolous one,
“Is it just me or do you think Jeff from Big Brother 11 usa looks alot
like the actor Jason Bateman?”
How can we better design social Q&A sites so that they are more
sensitive to users’ needs and constraints? In this article, I will discuss
the strengths and weaknesses of three types of solutions.
The most direct solution is to allow question askers and answerers to
share contextual information. If these communicating parties can be
more informed about each others’ needs and constraints, they may be
able to make better decisions. These solutions have been explored in
general communication domains.
Much research on media spaces in the 1990s explored how to use
technology to improve awareness of remote collaborators. Colleagues
can glance into another’s workspace and engage opportunistically [ 1,
2], even though they may be working thousands of miles apart.
More recently, different types of status update mechanisms have
been incorporated into everyday communication. These updates allow
users to share location and activity information, as well as business-related project updates. With this meta-level information, askers can
potentially target answerers who are available, and answerers can better infer the needs of the askers.
In my own work, I have explored the use of instant messaging tags
to provide communication initiators with a way to signal their information needs [ 3]. People often use text tags in email subject lines to
denote the type of email they are sending. By having a programmable
set of tags for communication, additional services can be automatically triggered. For example, using the tag [15m] in a message indicates
a level of urgency—a response within 15 minutes is desired. An automated reminder can be triggered when the time is up.
This type of support can be easily extended to online Q&A services.
Question askers can provide additional information on how urgently
they need the information, and answerers can then respond accordingly.
The main advantage of this solution is that it is straightforward and
intuitive to the users; people are accustomed to using available contextual information such as gestures, body positioning and verbal statements
to handle face-to-face requests. However, there are two major problems
with this type of solution. First, sharing contextual information is only
beneficial if askers and answerers have an incentive to respect each others’ constraints and needs. We may be able to expect this from communicating partners who have existing social relationships, but we cannot
expect this when communicating with strangers. Consider the case of the
spammer, who may send spam regardless of how busy the answerers are.
Second, full information disclosure has potential privacy problems. Not
everyone is willing to offer full-disclosure, especially to strangers.
Instead of disclosing relevant information and relying on askers and
answerers to make the proper decision on how to handle question
requests, the second type of solution uses computing mechanisms to
mediate social Q&A. This includes using collaborative filters to reduce
spam and minimize unwanted requests, as well as utilizing social networks to target the questions to more appropriate answerers, as is
done with Aardvark and Answer Garden.
But “intelligent” solutions can also leverage machine learning models to help determine when and how to target answerers for impending information requests. Using sensors placed in the environment,
machine learning models can predict the answerers’ interruptibility
(see, for example, Fogarty [ 4]), which can then be used to prevent
interruptions at inopportune moments. Additional models have been
built to examine the cost of deferring communication, which can be
used to improve the mediated decision of when to interrupt answerers (see, for example, Horvitz, Jacobs, and Hovel [ 5]).
Intelligent mediation designs have the potential to reduce request
overload and minimize interruption costs for real-time Q&A services.
Furthermore, individuals’ privacy may not be violated as contextual