|
|  |
|
in-cites,
November 2004
Citing URL: http://www.in-cites.com/scientists/JonHerlocker.html
|
|
|
|
An
interview with:
Dr. Jon Herlocker |
|
|
|
 ccording
to an analysis of the ISI
Essential
Science Indicators
Web product, Dr. Jon Herlocker’s work has recently entered
the top 1% in terms of total citations in the field of
Computer Science. His current citation record in this field
includes 3 papers cited a total of 107 times to date. Dr.
Herlocker is an Assistant Professor in the Department of
Computer Science at Oregon State University. In the interview
below, he talks about his highly cited work.
|
Why do you think your work is highly cited?
|
“The concept behind collaborative filtering is that a community of people shares the effort of filtering a large quantity of
information—separating the 'good' information from the
'bad' information.”
|
|
When I was a Ph.D. student at the University of Minnesota, I
worked with a research group called GroupLens Research. We were one
of the first groups of researchers to deploy automated/collaborative
filtering technology and publish empirical studies of that
technology. The concept behind collaborative filtering is that a
community of people shares the effort of filtering a large quantity
of information—separating the "good" information from
the "bad" information. Whenever one person
"experiences" some item of information, they share their
opinion of that item with the collaborative filtering system. The
collaborative filtering system then matches together people who seem
to have similar needs, based on correlations in their opinions.
Recommendations can then be transferred between people of similar
interest—both recommendations of what to see and what to avoid.
Our research group was one of the original innovators—not long
after, collaborative filtering went commercial at places like
Amazon.com & Netflix.com.
In my work, both with the GroupLens Research group at Minnesota
and now leading my own group at Oregon State University, I have the
goal of trying to ensure that every new innovation we come up with
is tested on at least 100 real people. This forces us to address the
practical issues of implementation, deployment, and user acceptance.
I believe that because we validate our work in such realistic
situations, across such a wide variety of metrics, our findings are
often very compelling to others.
What are the circumstances which led you to your work?
Mostly a coincidence. The original idea for the collaborative
filtering project at Minnesota came from a professor there, John
Riedl, and his associates. At the time (1994-1995), I was working on
something entirely different (highly adaptive user interfaces for
on-demand streaming media), but I helped them implement the first
Usenet news collaborative filtering trial. I found the idea very
compelling, so when the lead Ph.D. student left to start a company,
I took over student leadership of the collaborative filtering
research group and I never looked back!
How would you describe the significance of this work for your
field?
The work of my colleagues and I really integrates the algorithmic
and learning aspects of collaborative filtering and the human
aspects. The algorithms take large amounts of data and predict what
each user will or won't be interested in. However, improvements in
algorithms can be meaningless if the user interface is cumbersome,
or sociological effects prevent the acceptance of that interface. I
believe that our consideration of the interaction between
algorithmic and sociological factors makes our conclusions more
practically applicable.
Where do you see this research going 10 years from now?
Primarily, the major impact of future research will come from
collaborative filtering being integrated into more and more domains.
In particular, we will see it integrated into document and web page
search engines. Currently, collaborative filtering has primarily
just penetrated the books and entertainment market. There are
research challenges in web page and document search because users'
information needs change much more dynamically in those areas than
in entertainment. Hopefully, 10 years from now, we will have figured
out issues such as how to handle evaluating items on multiple
dimensions, how to better models users as combinations of different
sub-tastes, and how to design collaborative filtering systems for
domains where there is much more risk in acting on a decision, among
other things.
What lessons would you draw from your work to share with the
next generation of researchers?
Testing your research ideas with real implementations on real
people will help ensure that you are investigating the research
questions that really matter! In cases where that is impractical, be
honest with yourself about whether finding the answer to a research
question will really have practical impact on the world. If you have
to work hard to justify it, there's probably a better question to
pursue.
Jon Herlocker, Ph.D.
Department of Computer Science
Oregon State University
Corvallis, OR, USA
| Dr.
Jon
Herlocker's
most-cited paper with 107 cites to date: |
|
Konstan, JA, et al., "GroupLens: Applying collaborative filtering to Usenet news,"
(Commun. ACM 40[3]: 77-87, March 1997). |
|
Source:
ISI
Essential Science Indicators |
|
|
|
|
in-cites, November 2004
Citing URL: http://www.in-cites.com/scientists/JonHerlocker.html
|
|
cites
