122 August 2000/Vol. 43, No. 8 COMMUNICATIONS OF THE ACM

STEV

E A

DLE

R

COMMUNICATIONS OF THE ACM August 2000/Vol. 43, No. 8 123

Personalizationon the Netusing WebMining�

Maurice D. Mulvenna, Sarabjot S. Anand, andAlex G. Büchner, Guest Editors

On the Internet, we have experienced massive growth in systems

that can personalize content delivered to individual users. The

science behind personalization has undergone tremendous

changes in recent years, yet the basic goal of personalization systems remains

the same: to provide users with what they want or need without requiring

them to ask for it explicitly. Personalization is theprovision to the individual of tailored products, ser-vices, information or information relating to prod-ucts or service. It is a broad area, also coveringrecommender systems, customization, and adaptiveWeb sites.

Three aspects of a Web site affect its utility in pro-viding the intended service to its users. These are thecontent provided on the Web site, the layout of theindividual pages, and the structure of the entire Website itself. The relevance of each of the objects com-prising a Web page to the users’ needs will clearlyaffect their level of satisfaction. The structure of theWeb site, defined by theexistence of links betweenthe various pages, restrictsthe navigation performed bythe user to predefined pathsand therefore defines theability of a user to access rel-evant pages with relativeease. However, the defini-tion of relevance is subjec-tive. It is here that there is apotential mismatch betweenthe perception of what theuser needs, on the part ofthe Web site designer, andthe true needs of users. This may have a majorimpact on the effectivenessof a Web site.

Personalization technol-ogy involves software thatlearns patterns, habits, andpreferences. On the Inter-net, its use is primarily in systems that support e-business. Personalization works in this contextbecause it helps users to find solutions, but perhapsmore importantly, it also empowers e-businessproviders with the ability to measure the quality ofthat solution. In terms of the fast emerging area ofCustomer Relationship Management (CRM), per-sonalization enables e-business providers to imple-ment strategies to lock in existing customers, and towin new customers.

Daniel E. O’Leary from the University of South-ern California coined the phrase “AI Renaissance” in1997,1 to describe how artificial intelligence (AI) canmake the Internet more usable. Personalization tech-nology is part of that renaissance. In parallel with

the academic progress covered in this special section,the commercial world is witness to unprecedentedgrowth in personalization technology companies. Itis sometimes difficult to find a commonality in tech-nology foundation that spans the breadth of com-mercial product offerings and global academicefforts in personalization, as well as the broad crosssection of emerging efforts in digital markets.

Initial attempts at achieving personalization onthe Internet have been limited to check-box person-alization, where portals allow the user to select thelinks they would like on their “personal” page. How-ever, this has limited use since it depends on the

users knowing beforehandthe content of interest tothem. Arguably, collabora-tive filtering was the firstattempt at using AI forachieving personalizationin a more intelligent man-ner. This allows users totake advantage of otherusers’ behavioral activitiesbased on a measure of sim-ilarity between them.These techniques requireusers to divulge some per-sonal information on theirinterests, likes and dislikes,information that manyWeb users would not nec-essarily wish to divulge. Analternative is observationalpersonalization, whichattempts to circumvent theneed for users to divulge

any personal information. The underlyingassumption in this approach is that hidden withinrecords of a user’s previous navigation behavior areclues to how services, products, and informationneed to be personalized for enhanced Web interac-tion.

In our view, there are three principal componentsto observational personalization: analytics, represen-tation, and deployment. Web mining provides thetools to analyze Web log data in a user-centric man-ner such as segmentation, profiling, and clickstreamdiscovery. The knowledge mined by using thesetools is increasingly being represented using W3Cstandards such as XML and the deployment of theknowledge on Web servers may be carried outthrough personalization or recommender systems.

Perhaps for the first time since the beginning ofO’Leary’s AI renaissance, research and commercialgoals are identical. These goals are high perfor-

124 August 2000/Vol. 43, No. 8 COMMUNICATIONS OF THE ACM

1See O’Leary, D. The Internet, intranets, and the AI Renaissance. IEEE Computer,(Jan. 1997).

Personalization will beubiquitous in digitalinteractive devices,

from handheld computers throughmobile telephony

devices to digital TV.

mance, high quality analytics, flexible and novel rep-resentational schema, and real-time application ordeployment of the represented knowledge on Webservers, WAP servers, and so forth.

This special section contains four articles cho-sen to reflect various aspects of personalization onthe Net using Web mining. Spiliopoulou providesa rationale for why Web log data should be mined.The effectiveness of a Web site in providing userswith the content they need in the most optimizedmanner is the key to retaining them. Traditionalfocus group based methods for evaluating theeffectiveness of a Web site’s structure are expensiveand difficult to implement because little is knownabout the underlying population of users of thesite and their needs. Spiliopoulou describes aprocess by which mining for navigational patternsmay be used to gain insight into a Web site’s usageand optimality with respect to its current user population.

Cingil, Dogac, and Azgin describe the need forinteroperability when mining the Web and how thevarious W3C standards can be used to achieve per-sonalization applications. In particular, the articlelooks at how the recent data exchange, metadataand privacy standards from the World Wide WebConsortium (W3C), namely, Extensible MarkupLanguage (XML), Resource Description Frame-work (RDF), and Platform for Privacy Preferences(P3P) may be used to support personalizationactivities.

Mobasher, Cooley, and Srivastava provide aframework for mining Web log files to discoverknowledge for the provision of recommendations tocurrent users based on their browsing similaritieswith previous users. The process for discovering suchknowledge includes gathering and preprocessing thedata necessary for discovering user behaviors, appli-cation of data mining techniques to discover usagepatterns, and aggregation and filtering of the datamining results in order to create decision rules forcustomizing Web site content based on an individualuser’s behavior.

Perkowitz and Etzioni address personalization as aprocess that adapts an Internet site through the auto-mated generation of index pages for the Web site.The article explores adaptive Internet sites: sites thatautomatically improve their organization and pre-sentation by learning from visitor access patterns.Adaptive Web sites mine the data buried in Webserver logs to produce more easily navigable Websites.

We hope you find these articles interesting anduseful, and that they show the potential and diverse

uses available for personalization systems. In Febru-ary 1999, the Gartner Group stated that, “matchingdirect or inferred reader requests through contentpersonalization will be the most dramatic develop-ment in the Internet…through 2002, and will helpdifferentiate the Web as a new medium.” Clearly thisis an important technology, and is being applied inmany Internet systems in use today. As digital mar-kets converge, personalization will be ubiquitous indigital interactive devices, from handheld computersthrough mobile telephony devices to digital TV.

Little attention to date has been paid on evaluat-ing the success of Web sites in terms of their utilityto the users. Server statistics provided by Web loganalysis tools provide metrics for evaluating the suc-cess of the server in serving pages to users, however,no insights are available regarding how useful thecontent provided was to the user. Evaluation of theeffectiveness of a Web site must be based on thebusiness goal of the e-business and these metricswould provide the basis of evaluating the success ofpersonalization activities. Web Intelligence toolsbased on Web mining have an important role to playin the development of these e-metrics. ACM’s KDDconference in Boston (Aug. 20–23) will feature e-metrics and the role of Web mining in personalization.

The success of personalization on the Webdepends on the ability of the personalization com-munity in promoting responsible use of the technol-ogy by e-businesses. (“Responsible,” like “relevant,”is a subjective term.) A recent survey2 of Web userssuggests over 70% of Web users find online solici-tation from a Web site a hindrance rather thanhelpful and over 50% of Web users are worriedabout their privacy.3 How the providers of person-alized services can manage to tread the fine linebetween personalization and personal intrusionand control the hype around it will determine thefuture of this field. Technologically, the scene is set,for the first time, for a mutually equitable relation-ship to develop between businesses and their customers.

Maurice D. Mulvenna (maurice@mineit.com) is the chief executive officer of MINEit Software Limited in Belfast, Northern Ireland.Sarabjot S. Anand (sarab@mineit.com) is the chief technologyofficer at MINEit Software Limited in Belfast, Northern Ireland.Alex G. Büchner (alex@mineit.com) is the chief developmentofficer at MINEit Software Limited in Belfast, Northern Ireland.

© 2000 ACM 0002-0782/00/0800 $5.00

c

COMMUNICATIONS OF THE ACM August 2000/Vol. 43, No. 8 125

2See www.personalization.org.3These are users that always read the privacy statement on a Web site before divulgingany personal information.