First Monday

Search engine personalization: An exploratory study

Abstract
Search engine personalization: An exploratory study by Yashmeet Khopkar, Amanda Spink, C. Lee Giles, Prital Shah, and Sandip Debnath
Web search engines are beginning to offer personalization capabilities to users. Personalization is the ability of the Web site to match retrieved information content to a user's profile. This content can be set explicitly by the user or derived implicitly by the Web site using such user profile information as zip code, birth date, etc. In this paper we report findings from a study qualitatively and quantitatively assessing the current state of personalization on 60 search engine Web sites and the personalization features available. We examined: (1) how many search engines Web sites currently offer personalization features; and, (2) the type of features that can be personalized. Findings show that: (1) eight (13 percent) of the 60 search engines, including Yahoo, AOL, Lycos, Excite and Netscape, enabled some level of personalization; and, (2) personalization features are largely related to e-mail, business and financial information, searching of a reference tool, such as yellow pages, entertainment listings, sports, and news headlines. The breadth and depth of personalization features varied across search engines, with a mean number of two personalization features per site. "My Yahoo" had the most extensive personalization feature capability. Our findings show that despite the high level of interest in Web personalization, most search engine Web sites currently offer no or limited personalization features.

Contents

Introduction
Related studies
Research questions
Research design
Results
Discussion
Conclusion

 


 

++++++++++

Introduction

As more search engines on the World Wide Web become available, a "personalized one-stop solution" is being discussed as a competitive feature to attract users. Some search engines offer personalization and others provide customization, or both. In personalization, a user-created profile decides the personalizable solution, whereas in customization, the user is allowed to select from a predefined set of solutions. Though personalization seems to be a simple concept to understand, it is not easy to implement. There are problems since personalization needs vary; one size does not fit all. Thus it's difficult to provide a personalization solution that is complete for each user. Search engines can provide a generic set of personalizable features and hope that the users are satisfied with them.

Another important issue is accessibility: How accessible are these personal features? As the user accesses more layers in their personalized Web site, the depth of personalization increases and studies show less chance for use. Snowberry et al. (1983) showed that error rates increased from four percent to 34 percent as menu depth increased from a single level to six levels for a GUI.

Web user model generation depends on the willingness of the users to provide information about themselves and their awareness of self-needs. A user who visits a real estate Web site to find an apartment may know what kind of an apartment she wants to live in and be willing to state her preferences explicitly. Thus, the user can build a model. Many Web users prefer not to provide personal information if they do not benefit from it immediately. A search engine user may prefer not to spend time creating a profile while searching for information for a research paper. In this case, the system has to construct the user model by inferring knowledge from the user's behavior and exploiting user feedback about the relevance of documents implicitly.

User interests may change over time. To reflect the true state of an individual's preferences, a user model should be dynamic. The user-modeling component has to be capable of deriving information about the user by monitoring her activities. Even if the system can capture changes in user interests, there may be a delay between the occurrence of change and its realization by the system. Therefore, uncertainty exists and has to be included in the user-modeling task. Machine learning and decision-theoretic inference methods are being explored for this purpose.

 

++++++++++

Related studies

Web personalization

There is a growing interest in providing automated personalization processes on Web sites (Barrett et al., 1997; Hirsh et al., 2000). Currently Web sites rely heavily on user inputs for a personalization solution. Mobasher et al. (2000) propose a general architecture for automatic Web personalization. The architecture attempts to automate the personalization process by tracking the user's interest from the Web server logs. Perkowitz and Etzioni (2000) addressed this problem by designing an adaptive Web site that relies heavily on the user's navigation pattern and tries to anticipate the user's need based on his past navigation history. Barrett et al. (1997) describe the WBI (Web Browser Intelligence) architecture for personalizing Web sites. The previous two approaches rely on a server for the personalization process, but the WBI architecture can be used on the client side, mid-stream or server side. Pretschner and Gauch (1999) discuss the applications for personalization that exists on the Web and the use of profiles in the personalization process. Few studies have surveyed the nature and extent of search engine Web sites that include personalization features.

There has been a growing interest in making the personalization process completely automated. Presently the Web sites rely heavily on user's inputs for presenting them a personalizable solution. Manber et al. (2000) discuss the applications for personalization that exist on the Web and the use of profiles in the personalization process.

User modeling

Constructing and managing accurate and comprehensive user models is of great importance for Web and e-commerce applications (Allen et al., 1998). User's homepage (Muller, 2000), favorites, and history files (Moukas and Maes, 1998) are some examples implicit user modeling systems have used as a source to generate the initial model of the user. Regardless of the method by which it is generated, a user model has to be capable of adapting to the changes in the user's interests and needs. The implicit user models can therefore be constructed incrementally, rather than generating an initial model and updating it (Horvitz et al., 2000). Stereotype profiles have also been used as starting point and modified to fit the actual user (Fink et al., 1998).

The electronic commerce community has recognized the possibility of turning Web surfers into customers through personalized marketing strategies. The importance of user modeling in this user-targeted market place is invaluable. A considerable number of systems for personalization on the Web, with different capabilities, are being advertised. The personalization tools that were most talked about during the writing of this report include Groups Lens (Net Perceptions at http://www.netperceptions.com), predicting user's interests based on collaborative filtering techniques, and explicit/implicit ratings by the user.

Based on a review of the related studies we determined that research was needed to assess the current level of personalization capability available on search engine Web sites.

 

++++++++++

Research Questions

Our research goals were to determine how many and how much of search engine sites are personalizable. This work reports findings that determine:

  1. How many search engine Web sites offer personalization features?
  2. What features can be personalized?
  3. How accessible are the personalizable features?

We qualitatively and quantitatively analyzed the existence and extent of personalization features in 60 search engine sites. At these sites we explored personalizable options, the presence of a particular feature across Web sites, and the depth of a personalization feature, i.e. the number of minimal links the user need to reach a personalization feature.

 

++++++++++

Research design

We examined 60 search engine sites taken from SearchEngines.com (at http://www.searchengines.com/) for personalization features listed in Figure 1 below.

 

Figure 1: Search engine sites.

AOL.com fast search (alltheweb.com) pepesearch.com
4anything.com gogettem.com quango.com
about.com google.com ragingsearch.altavista.com
aeiwi.com goto.com searchbug.com
allsearchengines.com hotbot.com searchenginecolossus.com
altavista.com hotlinks.com searchengineguide.com
Askjeeves.com hotrate.com searchiq.com
backflip.com infogrid.com searchitall.hypermart.net
blink.com inktomi.com searchking.com
britannica.com iwon.com searchpower.com
diabolos.com lastminutesearchengine.com skworm.com
Directhit.com links2go.com sportula.com
directoryguide.com looksmart.com sunsteam.com
disinfo.com lycos.com top9.com
dmoz.org msn.com topclick.com
efind.com nbci.com webcrawler.com
entireweb.com netflip.com webwombat.com
epilot.com netscape.com wherewithal.com
euroseek.com northernlight.com yahoo.com
excite.com octopus.com zeal.com

 

We quantitatively analyzed the absence or presence of personalizable features for each search engine site. Below are the units of analysis that we used in our study:

Personalizable Feature — a feature whose content can be edited, or if it is a general feature that can be added or removed on a personalized page, or a link to it exists from an editable section.

Extent of Personalization — the extent to which a search engine site allows the user to personalize the content, including the number of personalization features the user is allowed to manipulate to benefit their personalization process.

Depth of Personalization — the number of levels in the hierarchy (Paap and Cooke, 1997). A feature may be accessed on the personalized page or may be accessed by following a link to the next level. This traversal to levels gives us the depth of a feature.

Accessibility — feature "A" is said to be more accessible than a feature "B" if the depth of "A" is lesser than "B".

Featurization — the ability to provide the features that support personalization.

Note that if a feature is marked absent in the table of personalizable features, then it does not necessarily mean that the feature is absent. It only means that the feature is not personalizable. If e-mail was the only feature that was provided, then we have not counted it as a personalizable feature.

From 1 to 15 May 2001, we personalized every search engine site that included personalization features to their fullest extent. This involved registering at each of the personalized sites, logging into them, and adding each of the personalization features by going to add\edit content page. If a sub-feature existed for that particular feature we also tried to personalize it. For example, if we personalized a Stock Portfolio personalization feature, we added the Stock Portfolio personalization feature to the main page. The Stock Portfolio then was edited and the Ticker Symbols were added to create our personalized stock portfolio. In this way we personalized all the personalizable features associated with the search engine sites. For future study, copies of the search engine sites that we personalized were documented and are available. Some personalization features may have changed after our survey due to the dynamic nature of the Web. Some personalization features may not be directly accessible from the main page, but are accessible from another level below the main page. Some search engine sites provide personalization but also give default personalization features that cannot be edited.

We tabularized the data in tables where the columns are the search engines sites and the rows are the personalizable features. A "1" denotes that the feature is personalizable and a "0" denotes that the feature is not personalizable. We also compared graphically in three dimensions the depth of the features. Not all the features explored have been listed.

The next section of the paper provides the results of our study.

 

++++++++++

Results

Extent of personalization

We found that eight (13 percent) of the 60 search engine sites included some personalizable features, including Yahoo, MSN, Lycos, NBCi, Excite, AOL, Webcrawler, and Netscape.

Our analysis of search engine sites identified seven broad categories of personalizable features: Headlines, Message Center, Personal Finance, Tools, Types of Searches, Entertainment, and, Sports. All eight search engine sites have each of these major categories of personalization. It should be noted that other personalizable categories existed, such as "Family".

Each category is explained below:

There were cases where a particular personalization feature such as "Sports Scoreboard" was listed in Yahoo's Sports section and Lycos had a parallel feauture in its Interests section; we assigned the Scoreboard in the Sports section.

The next section of the paper describes in detail our findings in each personalization category.

Message Center category

Table 1 shows the most common message center personalization features.

 

Table 1: Message Center categories.

 
E-mail 1 1 1 1 1 1 0 1 7 87
Chat 1 1 1 1 1 1 1 0 7 87
Messenger 1 1 1 0 1 1 1 1 7 87
Calendar 1 1 1 1 1 1 1 1 8 100
Address Book 1 1 1 1 1 1 1 1 8 100
BriefCase 1 1 1 0 0 0 0 1 4 50
NotePad 1 1 0 1 1 0 1 0 5 62
Web Communities/Clubs 1 1 1 1 0 1 0 0 5 62
My Web Site 1 1 1 1 0 1 0 1 6 75
Videos 0 0 1 0 0 0 0 0 1 12
Photo Album 1 1 1 0 1 0 1 0 5 62
Message Board 1 1 1 0 0 1 0 1 5 62
Personals 1 1 0 1 0 1 1 1 6 75
Total sub-personalization features 12 12 11 8 7 9 7 8    
Percentage of sub-personalization features in a search engine 92 92 84 61 53 69 53 61    

 

 

The most common personalization features of a Message Center are: E-mail, Chat, Messenger, and Address book. The "Brief Case" is an interesting personalization feature that stores the user's files — Yahoo, MSN, Lycos and Netscape provide this personalization feature. Lycos was the only search engine Web site that allowed the upload of Videos. The Message Center contained the largest number of personalizable sub-features per search engine site, a mean of 71 percent. Yahoo and MSN have the most extensive Message Center personalization features. The Netscape Personals section is connected to the Matchnetwork.com site and a registration fee is required to access it. Registration for the Personals section in the remaining sites (where it appears) is free. The depth of each personalization feature in the various Message Centers is shown in Figure 2.



Figure 2: Message center personalization feature accessibility.

Greater depth means that more links had to be traversed to reach the personalization feature. Yahoo provides the most accessibility for Message Center. Personalization features such as E-mail and Calendar are the most accessible, whereas My Web Site, Web Communities, and Personals are the most difficult to access.

Tool Category

Table 2 shows the personalization features in the tool category.

 

Table 2: Tools personalization features: Existence of sub-personalization features for eight search engine sites.

 
Bookmarks\Link 1 1 1 1 1 0 1 1 7 87
My Auto 1 0 0 0 0 0 0 0 1 12
Saved Searches 1 0 0 0 0 0 0 0 1 12
Package Tracker 1 0 0 0 0 0 0 0 1 12
Concert Tickets 0 1 0 0 1 0 1 0 3 37
My Miles 0 0 0 0 1 0 1 0 2 25
Calorie Calculator 0 0 0 0 1 0 1 0 2 25
Quick Click 0 0 0 1 0 0 0 0 1 12
My Recommendations 0 0 0 1 0 0 0 0 1 12
Fortune Teller 0 0 1 0 0 0 0 0 1 12
Traffic 1 0 1 0 0 1 0 0 3 37
Lottery 1 0 1 1 1 0 1 0 5 62
Recipe Finder 0 0 0 0 0 1 0 0 1 12
My Government 0 0 0 0 0 1 0 0 1 12
Total sub-personalization features 6 2 4 4 5 3 5 1    
Percentage of sub-personalization features in a search engine 42 14 28 28 35 21 35 7    

 

 

Bookmarks, Saved Searches, and Package tracker are unique to Yahoo! Quick Click and My Recommendations are unique to NBCi. Recipe Finder and My Government are unique to AOL. The Fortune Teller is unique to Lycos. Most unique personalization features are accessible from the main page.

Figure 3 shows the accessibility of tools personalization features.



Figure 3: Tools personalization feature accessibility.

Many tool personalization sub-features were not very deep. The deepest personalization feature was Excite's Concert Ticket options.

Entertainment Category

Table 3 shows the personalization features in the entertainment category.

 

Table 3: Entertainmnet personalization features: Existence of sub-personalization features for eight search engine sites.

 
Comics\Puzzles 1 0 1 0 1 1 1 0 5 62
Horoscopes 1 1 1 1 1 1 1 1 8 100
TV Listings 1 0 1 1 1 1 1 1 7 87
Movie Show Times 1 0 0 1 1 1 1 1 6 75
Album Releases 1 0 0 0 0 0 0 1 2 25
Movie Releases 1 0 1 0 1 0 1 1 5 62
Top Ten Videos 1 0 1 0 1 1 1 1 6 75
Top Ten Books 1 0 1 0 0 0 0 0 2 25
Survivor Watch 1 0 0 0 0 0 0 0 1 12
Free Games 0 0 1 0 1 1 1 0 4 50
Music Downloads 0 0 1 0 1 0 1 1 4 50
PC Downloads 0 0 1 1 0 1 0 1 4 50
Radio 0 0 1 0 1 1 1 1 5 62
Total sub-personalization features 9 1 10 4 9 8 9 9    
Percentage of sub-personalization features in a search engine 69 7 76 30 69 61 69 69    

 

 

The Horoscope sub-personalization feature is the most common Entertainment feature across all search engine sites. Lycos and NBCi allow only one horoscope star sign to be displayed and WebCrawler and Excite allow two star signs to be displayed. The Entertainment section has less average personalizable sub-features per search engine than the Message Center category. In this section Lycos has the highest number of sub-personalization features and MSN has the lowest number of sub-personalization features with only one. This is not because MSN does not have these personalization features, but that the MSN features are not personalizable.

Figure 4 shows the accessibility of Entertainment personalization features.



Figure 4: Entertainment personalization feature accessibility.

Figure 4 shows that search engines such as Netscape, Lycos and Excite provide the most accessibility for Entertainment. Features such as Free Games and TV Listings are the most accessible. Yahoo! Provides most sub-features at a level 1 depth.

News and Headlines

Table 4 shows the most common personalization features in the News and Headlines category.

 

Table 4: Most common personalization features in the News and Headlines category.

 
Top News\Stories 1 1 1 1 1 1 1 1 8 100
World\International 1 1 1 1 1 1 1 0 7 87
U.S. News 1 1 0 1 1 1 1 1 7 87
Sports 1 1 1 1 1 1 1 1 8 100
Politics 1 1 1 1 1 1 1 1 8 100
Health 1 1 1 1 1 1 1 1 8 100
Science 1 1 1 1 1 1 1 1 8 100
Business 1 1 1 1 1 1 1 1 8 100
Technology 1 1 1 1 1 1 1 1 8 100
Entertainment 1 1 1 1 1 1 1 1 8 100
Local\Newspaper 1 1 1 0 0 0 1 0 4 50
Market\Stock News 1 1 0 1 1 1 1 1 7 87
Financial News 1 1 1 1 0 0 0 0 4 50
News Tracker 1 0 0 1 1 0 1 0 4 50
Odd News\Offbeat 1 0 1 1 1 1 1 1 7 87
News Lookup\Search 0 0 0 1 0 0 0 1 2 25
Living 1 1 0 1 0 0 0 0 3 37
Travel 1 0 0 1 1 0 1 0 4 50
Autos 1 0 0 1 0 0 0 0 2 25
Society 1 1 0 1 0 0 0 0 3 37
Kids and Family 0 1 0 1 0 0 0 0 2 25
Commentary 1 1 0 0 0 0 0 0 2 25
Total sub-personalization features 20 17 12 1 14 12 14 14    
Percentage of sub-personalization features in a search engine 91 77 54 86 63 54 63 63    

 

 

In the Headlines and News section all the Web sites have sub-sections like: Top News Stories, Sports, Politics, Health, Science, Technology and Entertainment. This shows that all of the Web sites feel that most of the user's interest in News can be classified into these categories. This section is placed on the main page for all the Web sites, with the Headlines and News section the most accessible personalization feature. Netscape and NBCi provide the News Lookup personalization feature that helps in searching news on a particular topic. This personalization feature facilitates access to the entire News database of the Web site to the user. Lycos provides Local News of the U.S. states and Yahoo! provides news of the major cities of the world. The mean percentage of sub-features in News and Headlines was 69 percent. This section is on the first page and hence the depth of the sub-feature is always one.

Personal Finance

Table 5 shows the most common personalization features in the Personal Finance category.

 

Table 5: Most common personalization features in the Personal Finance category.

 
Stock Portfolios 1 1 1 1 1 1 1 1 8 100
Currency Converter 1 0 1 0 0 0 0 0 2 25
Upgrades\Downgrades 1 1 1 0 0 0 0 0 3 37
Finance Vision\Insight 1 1 0 0 0 0 0 1 3 37
Retirement Planner 0 1 1 0 1 0 1 0 4 50
Market Talk\Chat 1 1 0 1 1 0 1 0 5 62
Market Indices 1 1 1 1 1 1 1 1 8 100
Quote Search 1 1 1 1 1 1 1 1 8 100
Financial Accounts 1 1 1 0 1 0 1 0 5 62
Bill Pay 1 1 1 0 1 0 1 0 5 62
Mortgage Monitor 1 1 1 0 1 1 1 0 6 75
Pay Direct 1 1 0 0 0 0 0 0 2 25
Total sub-personalization features 11 11 9 1 8 4 8 4    
Percentage of sub-personalization features in a search engine 91 91 75 33 66   66 33    

 

 

In the Personal Finance section the Stock Portfolio, Market Indices and Quote Search are the sub-features common to all search engine sites. The Currency Converter personalization feature is unique to Yahoo! and Lycos. Pay Direct is a unique feature in Yahoo! and MSN, which allows the user to pay through E-mail. Yahoo! and MSN have the maximum sub-features. The Personal Finance section has an average of 61 percent of sub-features per search engine site. The depth of each personalization feature in the Personal Finance is shown below.



Figure 5: Personal Finance personalization feature depth.

As seen from the graph, Stock Portfolios and Quote Search are the most accessible sub-features in the Personal Finance section. The sub-features of Financial Accounts, Bill Pay, Retirement Planner and Mortgage Monitor are the most difficult to access. If you look at Yahoo! and MSN we can see that they have almost all the personalization features. But Figure 5 brings out the difference in terms of accessibility. Yahoo! has most of the personalization features on the first page. MSN has most of the personalization features, but they are placed five or six levels deep. As seen from Figure 5 Yahoo! has most of the sub-features and is the most accessible with respect to the Personal Finance sub-features.

The next section of the paper discusses the search features that were personalizable for each search engine site.

Types of Searches

Table 7 shows the most common personalization features in the Type of Search category.

 

Table 7: Most common personalization features in the type of search category.

 
Yellow Pages 1 1 1 1 1 0 1 0 6 75
White Pages 1 1 1 1 1 0 1 0 6 75
Maps\Driving Directions 1 1 1 1 1 1 1 0 7 87
Flight Finder 1 1 1 1 1 1 1 1 8 100
Quote Search 1 0 1 1 1 1 1 1 7 87
Weather Lookup 1 0 0 1 1 1 1 1 6 75
Show Times\Ticket Lookup 1 0 0 0 1 1 1 1 5 62
Artist Search 0 0 1 0 0 0 0 1 2 25
News Look-up 0 0 0 1 0 0 0 1 2 25
My Sub Search 1 0 1 1 0 0 0 0 3 37
Software Search 0 0 1 1 0 1 0 0 3 37
Auction Search 1 0 1 1 0 0 0 1 4 50
Auto Search 0 0 0 1 0 0 0 1 2 25
Picture Search 0 0 0 1 0 1 0 1 3 37
Audio Search 0 0 0 1 0 1 0 1 3 37
Video Search 0 0 0 1 0 1 0 1 3 37
Recipe Finder 0 0 0 0 0 1 0 0 1 12
My Government 0 0 0 0 0 1 0 0 1 12
Total sub-personalization features 9 4 9 14 9 11 7 11    
Percentage of sub-personalization features in a search engine 50 22 50 77 50 61 38 61    

 

 

Maps and Driving directions, Quote Search, and Flight Finder are found in most of the search engine sites. NBCi has the maximum types of searches options.

Figure 6 shows the depth of each type of search feature.



Figure 6: Type of search personalization feature depth.

Yellow Pages, White Pages, Map and Driving directions, and Weather Look up are generally accessible on the main page.

Sports

Table 8 shows the most common features in the Sports category.

 

Table 8: Most common personalization features in the Sports category.

 
Sports Scoreboard 1 1 1 1 1 1 1 1 8 100
Fantasy Sports 1 0 1 1 0 0 0 0 3 37
Ski Report 1 0 0 1 1 0 0 0 3 37
Team Calendars 1 0 0 0 1 0 1 0 3 37
Buy Tickets 1 0 0 0 1 0 1 0 3 37
Total sub-features 5 1 2 3 4 1 3 1    
Percentage of sub-features in a search engine 100 20 40 60 80 20 60 20    

 

 

Yahoo! and Excite have the maximum number of sub-features. The Sports Scoreboard sub-feature is found in each search engine sites. The average sub-features in the Sports section are 50 percent.

Figure 7 shows the depth of each Sports personalization feature.



Figure 7: Type of search personalization feature depth.

Sports Scoreboards, Ski Report and Team Calendars are easily accessible. The Buy Tickets sub-feature in Yahoo! and Excite are difficult to access. Similarly Fantasy Sports is not accessible in the case of Lycos and NBCi.

Content Layout

Content layout describes how the personalizable features can be displayed on a page. Various content layout options are provided by the search engine sites ( Table 9).

 

Table 9: Content Layout: Existence of layout options for eight search engine sites.

  Number of pages Columns Drag 'n Drop Refresh rate
Yahoo! 6 2, 3 0 1
MSN 1 1 0 0
Lycos 4 2, 3 1 0
NBCi 3 2 0 1
Excite 1 2, 3 1 1
AOL 1 2 1 1
WebCrawler 1 2, 3 1 1
Netscape 1 3 1 1

 

 

For example, six pages in Yahoo!, four pages in Lycos, three in NBCi and for the rest of the Web sites only one page can be personalized. Drag 'n Drop allows the module to be rearranged on the personalizable page. Refresh rate is a personalization feature, which allows important information like Stock Portfolio to be updated automatically. There was also a page where the placement of the content could be specified. The placement ranged from a two-column to three-column page structure. Predefined color templates could also be applied to the Web sites. Lycos and NBCi provide predesigned templates for users who don't prefer extensive personalization. This feature can be useful, as the user then views an example of a personalized page.

 

++++++++++

Discussion

Our study shows that many personalization features lack accessibility and are not easy to personalize. For example, we found some personalizable features at search engine sites through their site map simply because we knew a similar personalization feature existed in another search engine. In this case, we had to trace the personalization feature back through the personalization page. Yahoo! seemed to provide the most effective integration of the personalizable modules with the Web site. Personalization can be readily determined by the user's profile, but it was quite evident from the study that the profile was used only for minor personalization tasks, such as registration, password retrieval, greeting the user, and giving the weather and horoscope. The user profile in all the cases did not play an active role in the personalization process.

In some cases users needed to login many times when accessing the personalized modules from the same search engine site as many modules are not well integrated. A close inspection of the personalization features being offered suggests that some search engines were trying to imitate personalization features provided by other search engines. In these cases where personalizable features existed, they were usually at the most inaccessible levels.

We were surprised to note that there was little or no personalization of search functions, an expected characteristic of search engine personalization. We suggest that the personalization process be taken to a new level: A level where the user does not to be actively involved with the personalization process. All that the user needs to do is to have an active profile file and when the user logs onto a Web site, the browser checks for that profile file as it checks for cookies. The profile file describes the user's interest and the levels at which the user wants a particular personalizable feature. Since the profile file is in a standardized format, Web sites would be able to provide the content according to the profile file. This would enhance the user's personalization process without their active involvement.

A general a user modeling component should be able to learn about the user from implicit sources, since the users are generally not willing to give information about themselves. Nevertheless, the user modeling component should also be able to let the user give explicit feedback or modify her model, since the user is the best source to learn about personal preferences. The choice of right model to represent the user and inference methodologies were also found to depend on the nature of the application. Bayesian networks may be a good method to use when the observed inputs of user information contain uncertainty. The ability of adapting both the conditional probability values and the structure of the network without direct user interaction is its advantage.

 

++++++++++

Conclusion

We are conducting further research into the personalization of search engines. Key issues include discerning what features are most effective to personalize and determining the best techniques for the development of personalization features. The knowledge of user's current needs and goals, or even better, what she will need or want in the future, is invaluable information for many application domains, especially where the user is a potential consumer. Although not restricted to Internet, its growing popularity for e-commerce has contributed to an increase in interest in user modeling. We plan to explore the use of integrated models for each user. Using more than one profile may make it possible to use all the advantages of different user modeling techniques, facilitating adaptation to the user's dynamic interests and at the same time simplifying the system's overall design. End of article

 

About the Authors

Amanda Spink and C. Lee Giles are from the School of Information Sciences and Technology at Pennsylvania State University. Yashmeet Khopkar, Prital Shah, C. Lee Giles, and Sandip Debnath are from Pennsylvania State Univesity's Department of Computer Science and Engineering.
E-mail: Correspondence should be directed to Amanda Spink at spink@ist.psu.edu.

 

Bibliography

C. Allen, D. Kania, and B. Yaeckel, 1998. Internet world guide to one-to-one Web marketing. New York: Wiley.

R. Barrett, P.P. Maglio, and D.C. Kellem, 1997. "How to personalize the Web," In: Proceedings of the ACM Conference on Human Factors in Computing Systems (CHI '97), pp. 75-82, and at http://www.almaden.ibm.com/cs/wbi/papers/chi97/wbipaper.html.

J. Fink, A. Kobsa, and A. Nill, 1998. "Adaptable and adaptive information access for all users, including the disabled and the elderly," New Review of Hypermedia and Multimedia, volume 4, pp. 163-188; and at http://www.isr.uci.edu/~kobsa/papers/1998-NRHM-kobsa.pdf.

H. Hirsh, C. Basu, and B.D. Davison, 2000. "Learning to personalize," Communications of the ACM, volume 43, number 8, pp. 102-106.

E. Horvitz, J. Breese, D. Heckerman, D. Hovel, and K. Rommelse, 2000. "The Lumiere project: Bayesian user modeling for inferring the goals and needs of software users," In G.F. Cooper and S. Moral (editors). Uncertainty in Artificial Intelligence: Proceedings of the Fourteenth Conference. San Francisco: Morgan Kaufmann, pp. 256-265; and at http://www2.sis.pitt.edu/~dsl/UAI/UAI98/Horvitz.UAI98.html.

U. Manber, A. Patel, and J. Robinson, 2000. "Experience with personalization on Yahoo!" Communications of the ACM, volume 43, number 8, pp. 35-37.

B. Mobasher, R. Cooley, and J. Srivastava, 2000. "Automatic personalization based on Web usage mining," Communications of the ACM, volume 43, number 8, pp. 142-151, and at http://maya.cs.depaul.edu/~mobasher/personalization/.

A. Moukas and P. Maes, 1998. "Amalthaea: An evolving multi-agent information fil-tering and discovery systems for the WWW," Autonomous Agents and Multi-Agent Systems, volume 1, pp. 59-68.

M.E. Muller, 2000. "OySTER: Web search as a playground for user modeling techniques," In: M.E. Muller (editor). Adaptivität und Benutzermodellierung in interaktiven Softwaresystemen. Institut für Semantische Informationsverarbeitung, Universität Osnabrück.

K.R. Paap and N.J. Cooke, 1997. "Design of menus," In: M.G. Helander, T.K. Landauer, and P.V. Prabhu (editors). Handbook of Human-Computer Interaction. New York: Elsevier, pp. 533-569.

M. Perkowitz and O. Etzioni, 2000. "Adaptive Web sites," Communications of the ACM, volume 43, number 8, pp. 152-158.

A. Pretschner and S. Gauch, 1999. "Personalization on the Web," Technical Report ITTC-FY2000-TR-13591-01, Information and Telecommunication Technology Center, University of Kansas, Lawrence; also at http://www4.in.tum.de/~pretschn/papers/personalization/index.html.

Searchenginewatch.com; at http://www.searchenginewatch.com, accessed 15 April 2001.

K. Snowberry, S.R. Pantlison, and N. Sisson, 1983. "Computer display menus," Ergonomics, volume 26, number 7, pp. 699-712.


Editorial history

Paper received 13 April 2003; accepted 4 June 2003.


Contents Index

Copyright ©2003, First Monday

Copyright ©2003, Yashmeet Khopkar, Amanda Spink, C. Lee Giles, Prital Shah, and Sandip Debnath

Search engine personalization: An exploratory study by Yashmeet Khopkar, Amanda Spink, C. Lee Giles, Prital Shah, and Sandip Debnath
First Monday, volume 8, number 7 (July 2003),
URL: http://firstmonday.org/issues/issue8_7/khopkar/index.html