Download Beyond Personalization 2005 - GroupLens Research PDF

TitleBeyond Personalization 2005 - GroupLens Research
File Size4.7 MB
Total Pages104
Table of Contents
                            Table of Contents
About the Workshop
Full Papers
	Crossing the Rubicon for An Intelligent Advisor
	Explaining Recommendations: Satisfaction vs. Promotion
	Identifying Attack Models for Secure Recommendation
	User-Specific Decision-Theoretic Accuracy Metrics for Collaborative Filtering
	Off-Topic Recommendations
	Item-Triggered Recommendation for Identifying Potential Customers of Cold Sellers in Supermarkets
	The Good, Bad and the Indifferent: Explorations in Recommender System Health
	Impacts of Contextualized Communication of Privacy Practices and Personalization Benefits on Purchase Behavior and Perceived Quality of Recommendation
	InterestMap: Harvesting Social Network Profiles for Recommendations
	What Affects Printing Options? - Toward Personalization & Recommendation System for Printing Devices
	P2P-based PVR Recommendation using Friends, Taste Buddies and Superpeers
	DynamicLens: A Dynamic User-Interface for Meta-Recommendation System
	Modeling a Dialogue Strategy for Personalized Movie Recommendations
	Behavior-based Recommender Systems for Web Content
Position Statements
	Who do trust? Combining Recommender Systems and Social Networking for Better Advice
	Recommender Systems Research at Yahoo! Research Labs
	A Multi-agent Smart User Model for Cross-domain Recommender Systems
	Personalized Product Recommendations and Consumer Purchase Decisions
	Toward a Personal Recommender System
	Beyond Idiot Savants: Recommendations and Common Sense
	Towards More Personalized Navigation in Mobile Three-dimensional Virtual Environments
	Issues of Applying Collaborative Filtering Recommendations in Information Retrieval
Document Text Contents
Page 1

Beyond Personalization 2005

A Workshop on the Next Stage of Recommender Systems Research

San Diego, January 9, 2005

In conjunction with the 2005 International Conference on
Intelligent User Interfaces (IUI 2005)

Edited by:
Mark van Setten

Sean McNee
Joseph Konstan

Page 52


to disclose


Quality of



+ +


Control of
own data





+ –



Figure 2: Proposed influence model

to have increased users’ trust and alleviated their privacy
concerns. This in turn led to more data disclosure.

The decision to buy a book was a significant step in our
experiment since at this point users revealed personally
identifiable information (name, shipment and payment
data) and risk that previously pseudonymous information
may be linked to their identities. We already reported above
that users indicate in surveys to refrain from shopping if
the are uncertain about the possible fate of their data. It
seems that the increased trust of users in condition “+expl”
due to contextualized privacy disclosure may have
contributed to more users opting to reveal their identities.

We have no direct explanation for the higher perceived
benefits from data disclosure in condition “+expl”. One can
speculate about positive transfer effects from higher
perceived privacy standards via higher trust.

Other characteristics of our experiment are also in
agreement with the literature. [14] found in their study of
consumer privacy concerns that “in the absence of
straightforward explanations on the purposes of data
collection, people were able to produce their own versions
of the organization’s motivation that were unlikely to be
favorable. Clear and readily available explanations might
alleviate some of the unfavorable speculation” [emphasis
ours]. [9] postulate that consumers will “continue to
disclose personal information as long as they perceive that
they receive benefits that exceed the current or future risks
of disclosure. Implied here is an expectation that
organizations not only need to offer benefits that consumers
find attractive, but they also need to be open and honest
about their information practices so that consumers […] can
make an informed choice about whether or not to disclose.”
The readily available explanations of both privacy practices
and personalization benefits in our experiment meet the
requirements spelled out in the above quotations, and the
predicted effects could be indeed observed.

Having said this, we would however also like to point out
that additional factors may also play a role in users’ data
disclosure behavior, which were kept constant in our
experiment due to the specific choice of the web retailer, its
privacy policy, and a specific instantiation of our proposed
interface design pattern. We will discuss some of these
factors in the following.

Reputation of a website. We chose a webstore that enjoys a
relatively high reputation in Germany (we conducted
surveys that confirmed this). It is well known that
reputation increases users’ willingness to share personal
data with a website (see e.g. [6, 12, 21]). Our high
response rates of 84% without and specifically 91% with
contextual explanation suggest that we may have already
experienced some ceiling effects. In a more recent version
of the experiment we therefore changed the name and logo
of the website to ones that had received a medium
reputation rating in the prior survey. We found indeed
similar effects of contextualized disclosures as at the
website with high reputation, but with smaller numbers for
data disclosure and purchases in both conditions. There was
no interaction between reputation and form of disclosure.

Figure 2: Suggested explanatory model

Stringency of a website’s data handling practices. The
privacy policy of the website that we mimicked is
comparatively strict. Putting this policy upfront and
explaining it in-context in a comprehensible manner is
more likely to have a positive effect on customers than
couching it in legalese and hiding it behind a link. Chances
are that this may change if a site’s privacy policy is not so

Permanent visibility of contextual explanations. In our
experiment, the contextual explanations were permanently
visible. This uses up a considerable amount of screen real
estate. Can the same effect be achieved in a less space-
consuming manner, for instance with icons that symbolize
the availability of such explanations? If so, how can the
contextual explanations be presented so that users can
easily access them and at the same time will not be
distracted by them? Should this be done through regular
page links, links to pop-up windows, or rollover windows
that pop up when users brush over an icon?

References to the full privacy policy. As discussed above,
privacy statements on the web currently constitute
important and comprehensive legal documents. Contextual
explanations will in most cases be incomplete since they
need to be short and focused on the current situation, so as
to ensure that users will read and understand them. For
legal protection, it is advisable to include in every
contextual explanation a proviso such as “This is only a
summary explanation. See <link to privacy statement> for
a full disclosure.” Will users then be concerned that a
website is hiding the juicy part of its privacy disclosure in
the “small print”, and therefore show less willingness to
disclose their personal data?

Additional user experiments will be necessary to obtain
answers or at least a clearer picture with regard to these

The work has been supported by the National Science
Foundation (grant DST 0307504), Deutsche Forschungs-
gemeinschaft (DFG grant no. GRK 316/2), and by
Humboldt Foundation (TransCoop program). We would
like to thank Christoph Graupner, Louis Posern and
Thomas Molter for their help in conducting the user
experiment described herein. The comments of the
anonymous reviewers are also appreciated.


Page 53


1. A Survey of Consumer Privacy Attitudes and
Behaviors, Harris Interactive, 2001.

2. Abrams, M., Making Notices Work For Real People.
in 25th International Conference of Data Protection
& Privacy Commissioners, (Sydney, Australia, 2003).

3. Ackerman, M.S., Cranor, L.F. and Reagle, J., Privacy
in E-commerce: Examining User Scenarios and Privacy
Preferences. First ACM Conference on Electronic
Commerce (Denver, CO, 1999), 1-8.

4. Berendt, B., Günther, O. and Spiekermann, S. Privacy
in E-Commerce: Stated Preferences vs. Actual
Behavior. Communications of the ACM (to appear).

5. Brodie, C., Karat, C.-M. and Karat, J. How Personali-
zation of an E-Commerce Website Affects Consumer
Trust. In Karat, C.-M., Blom, J. and Karat, J. eds.
Designing Personalized User Experience for
eCommerce, Kluwer Academic Publishers, Dordrecht,
Netherlands, 2004.

6. CG&I-R. Privacy Policies Critical to Online
Consumer Trust, Columbus Group and Ipsos-Reid,

7. Chellappa, R.K. and Sin, R. Personalization versus
Privacy: An Empirical Examination of the Online
Consumer’s Dilemma. Information Technology and
Management , 6 (2-3), 2005.

8. Cranor, L., Langheinrich, M., Marchiori, M., Presler-
Marshall, M. and Reagle, J. The Platform for Privacy
Preferences 1.0 (P3P1.0) Specification. W3C Recom-
mendation 16 April 2002,

9. Culnan, M.J. and Bies, R.J. Consumer Privacy:
Balancing Economic and Justice Considerations.
Journal of Social Issues, 59. 323-353.

10. Culnan, M.J. and Milne, G.R., The Culnan-Milne
Survey on Consumers & Online Privacy Notices:
Summary of Responses. In: Interagency Public Work-
shop Get Noticed: Effective Financial Privacy Notices,
(Washington, D.C., 2001).

11. Department for Trade and Industry. Informing
Consumers about E-Commerce, Conducted by MORI,
London: DTI, London, 2001. http://www.consumer.

12. Earp, J.B. and Baumer, D. Innovative Web Use to
Learn About Consumer Behavior and Online Privacy.
Communications of the ACM Archive, 46 (4). 81 - 83.

13. GartnerG2 Privacy and Security: The Hidden Growth

14. Hine, C. and Eve, J. Privacy in the Marketplace. The
Information Society, 14 (4). 253-262.

15. Kobsa, A. and Teltzrow, M. Contextualized Commu-
nication of Privacy Practices and Personalization Bene-
fits: Impacts on Users’ Data Sharing Behavior. In
Martin, D. and Serjantov, A., eds. Privacy Enhancing
Technologies: Fourth International Workshop, PET
2004, Toronto, Canada, Springer Verlag, Heidelberg,
Germany, to appear.

16. Kohavi, R., Mining E-Commerce Data: the Good, the
Bad, and the Ugly. in Seventh ACM SIGKDD Inter-
national Conference on Knowledge Discovery and
Data Mining, (San Francisco, CA, 2001), 8-13.

17. Lederer, S., Dey, A. and Mankoff, J. A Conceptual
Model and Metaphor of Everyday Privacy in Ubiqui-
tous Computing, Intel Research, 2002.

18. Palen, L. and Dourish, P., Unpacking “Privacy” for a
Networked World. in CHI-02, (Fort Lauderdale, FL,
2002), 129-136.

19. Roy Morgan Research. Privacy and the Community,
Prepared for the Office of the Federal Privacy
Commissioner, Sydney, 2001. http://www.privacy.

20. Spiekermann, S., Grossklags, J. and Berendt, B., E-
privacy in 2nd Generation E-Commerce: Privacy Pref-
erences versus Actual Behavior. in EC'01: Third ACM
Conference on Electronic Commerce, (Tampa, FL,
2001), 38-47.

21. Teo, H.H., Wan, W. and Li, L., Volunteering Personal
Information on the Internet: Effects of Reputation,
Privacy Initiatives, and Reward on Online Consumer
Behavior. in Proc. of the 37th Hawaii International
Conference on System Sciences (Big Island, HI, 2004).

22. van Duyne, D.K., Landay, J.A. and Hong, J.I. The
Design of Sites: Patterns, Principles, and Processes
for Crafting a Customer-Centered Web Experience.
Addison-Wesley, Boston, 2002.


Page 103

Issues of Applying Collaborative Filtering Recommendations in
Information Retrieval

Xiangmin Zhang
School of Communication, Information and Library Studies

Rutgers University
4 Huntington Street

New Brunswick, NJ 08901 USA
+1 732 932 7500 x 8229

[email protected]

My interest in the “Beyond Personalization 2005”
workshop is the application of collaborative filtering
recommendations in web information retrieval systems.

Collaborative filtering recommender systems are normally
found in the domains of movies, music, merchant products,
restaurants, and USENET newsgroup messages. One
common feature of these domains is that the item space is
relatively small, compared to the number of documents
available on the web. While the collaborative filtering
methods may work well for small item spaces, it is difficult
to be applied in the web information retrieval. Web
information retrieval is a domain with huge amount of
items/documents. In addition, the value of each item to the
user’s interest is up to a particular person who needs it.
Although there are some recommender systems for web
search, most of these systems are based on content-based
approaches, not the collaborative filtering approach [3, 4].

On the other hand, information retrieval is a natural domain
for the use of collaborative filtering recommendations. As
part of the scientific research process, collaboration in
information seeking is a common practice. People often
seek recommendations from colleagues or friends for the
needed information [2, 5]. As the number of digital
documents increases rapidly on the internet, the demand for
this collaboration becomes more and more urgent in order
to help people find relevant information. Collaborative
filtering recommendations should find a good fit in this

However, different from other domains, use of the
collaborative filtering for information retrieval tasks needs
to meet several challenges. In addition to the rating sparsity

and ramp-up problems typical in collaborative filtering
systems, as pointed out by many researchers [e.g., 1], we
also need to address the issues of 1) What to recommend
(relevant documents or the knowledge to find and identify
the relevant documents) and 2) what kind of users would
like to have recommended relevant documents.

We have recently worked on a project to explore the
effectiveness of collaborative filtering recommendations
for web search tasks. We built a pilot user interface system
that can display previous users’ relevant search results, as
well as the associated search queries. The system can use
one of the publicly available Internet search engines as the
information retrieval system and allow experimental
participants to search on a set of pre-defined search topics.
While the project is still on-going, part of our preliminary
results show that users prefer more other people’s queries
to their relevant judgments or relevant items. This implies
that when applying collaborative filtering
recommendations for web search tasks, the priority of what
to be recommended might be given to the search
knowledge of finding relevant documents, rather than the
documents themselves. The reason may be that the
relevance judgment (rating) is so subjective that people
trust only their own judgment. This is particularly true with
the trained information searchers. A detailed description of
this research will be presented elsewhere.

In summary, web information retrieval provides both an
opportunity and challenges for applying collaborative
filtering recommendations. The issues discussed in this
statement may be related to all topics concerned by this
workshop. I hope at the workshop I can learn other
people’s experiences and thoughts, and hope my research
can contribute to the filed as a whole.

Copyright is held by the author/owner(s).
Workshop: Beyond Personalization 2005
IUI'05, January 9, 2005, San Diego, California, USA

1. Burke, R. Hybrid recommender systems: Survey and

experiments. User Modeling and User-Adapted
Interaction 12, 4 (2002), 331-370.


Page 104

2. Karamuftuoglu, M. Collaborative IR: Toward a social
informatics view of IR interaction. Journal of the
American Society for Information Science, 49 (1998),

3. Konstan,J.A., Miller,B.N., Maltz, D., Herlocker,J.L.,
Gordon,L.R., and Riedl, J. GroupLens: Applying
collaborative filtering to USENET news.
Communications of the ACM, 40, 3 (March 1997), 77-

4. Montaner, M., Lopez, B. and Josep Lluis De La Rosa.
A taxonomy of recommender agents on the Internet.
Artificial Intelligence Review, 19, 4 (2003), 285-330.

5. Zhang, X. Collaborative relevance judgment: A group
consensus method for evaluating user search
performance”, Journal of the American Society for
Information Science and Technology, 53 (2002), 220-


Similer Documents