The Economics of Surfing
Eytan Adar and Bernardo A. Huberman
Xerox Palo Alto Research Center,
Palo Alto, CA 94304
Over the past four to five years we have seen the appearance of
several important WWW services for electronic commerce, of which a
relatively new instance are the portal sites. Some important examples
are Yahoo, Excite, and Lycos. Portals attempt to act as a starting
point for users on the web and therefore lead consumers to electronic
commerce activities, such as travel and consumer electronics. Portals
are in fact second generation services, and are essentially a
refinement of the web search engine service.
The business model of portal services consists of two parts. The first
one is to have the consumer buy goods directly from the portal or
through a partner site. The other strategy involves satisfying the
user's information needs locally and in the process present
advertising banners that result in direct revenues to the portal.
While this model seems to have worked for various companies, it is
highly paradoxical in the sense that it confronts the portal with a
dilemma. On the one hand, the service strives to generate better and
faster results for the users. However, if the results are presented at
the entry point, leading users to another site, there is an
opportunity cost for the provider. The user never travels deeply into
the site, thus missing the advertising and consumer goods that he
potentially could buy.
A recent heuristic solution to this problem resorts to the notion of
stickiness, whereby the provider attempts to keep consumers at their
site by displaying potentially attractive links that only point into
their site. This is much in line with standard marketing practices in
the physical world. An alternative solution, which we here explore,
attempts to capture the existence of different surfing patterns in
order to solve this dilemma. Instead of blocking surfing paths, one
increases the depth to which users surf when accessing an e-commerce
We show that depending on the domain of inquiry, users display
different and regular patterns of surfing users when accessing
different kinds of information goods or services. This difference can
be exploited in order to benefit information providers. We propose
mechanisms for implementing temporal discrimination in surfing by
dynamically configuring sites and versioning information services.
This is done by exploiting the fact that surfing patterns are
extremely regular and described by a universal law [Hu97]. We also propose mechanisms for extracting
consumer surplus by dynamically configuring sites and versioning
Experiments on surfing behavior.
While an aggregate demand curve seems indicative of different user
preferences in accessing information, it is of interest to know
whether or not specific surfing patterns depend on the information
being accessed. In order to answer this, we ran a set of experiments
on user logs provided by a large web portal and Excite.
The web portal data used in this study consisted of anonymized usage
logs for an eleven day period between September 14, 1998 and September
25, 1998. The data set contained over five million impressions, or
page views, generated by over five hundred thousand unique
visitors. The initial results of separating surfing patterns by high
level categories is illustrated in the figure below. These results
demonstrate the difference in surfing patterns exhibited by users
accessing information in different domains. The three curves
correspond to rate at which surfers complete their task in three high
In 1997 Excite provided the research community with
a small, anonymized usage log, containing over fifty thousand page
views from over ten thousand unique visitors. The log entry contained the unique
user id, a timestamp, and the query terms applied to search
engine. Each entry provides an
indication that the user viewed a single result page containing, by
default, ten matches. Similar
results to the portal category seperation were obtained for the Excite
query logs based on a separation of queries by adult versus non-adult
content. Users seeking adult information tended to display longer
surfing paths than any users seeking any other information. This In
both cases we observed a strong separation of the surfing behavior
depending on the domain surfed.
We propose three specific strategies that can be used to extract
maximum surf depths from consumers exploring a portal.
- Dynamic reconfiguration of the link structure of the site which
allows an extension of the path length depending on the domain being
surfed. These extends the notion of price discrimination for
traditional as well as information goods [Va99],
- A related technique of varying ranking schemes in a search engine
context similarly extends path lengths. The technique modifies
popular information retrieval ranking strategies [Sa68].
- Lastly, we discuss the proper timing for advertisement display,
leveraging the notion of the "seducible moment."
- [Hu97] Bernardo A. Huberman., Peter L. T. Pirolli, James
E. Pitkow, and Rajan M. Lukose, "Strong Regularities in World Wide
Web Surfing,"Science, April 3, 1997, 280(5360), pp. 95-97.
- [Sa68] G. Salton, and M. E. Lesk, "Computer Evaluation of Indexing
and Text Processing", Journal of the Association for Computing
Machinery, 1968, 15, pp. 8-36.
- [Va99] Hal R. Varian, "Market Structure in the Network Age", in
Understanding the Digital Economy, Department of Commerce,
Washington, DC 1999.
- [Wi93] Robert B. Wilson, Nonlinear Pricing, Oxford University
Further information on this work is available at: http://www.parc.xerox.com/istl/groups/iea/www/econsurf.html.