=Paper=
{{Paper
|id=Vol-1448/paper1
|storemode=property
|title=Keynote: Capturing User Interests for Content-based Recommendations
|pdfUrl=https://ceur-ws.org/Vol-1448/paper1.pdf
|volume=Vol-1448
|dblpUrl=https://dblp.org/rec/conf/recsys/Hopfgartner15
}}
==Keynote: Capturing User Interests for Content-based Recommendations==
https://ceur-ws.org/Vol-1448/paper1.pdf
Capturing User Interests for Content-based
Recommendations
Frank Hopfgartner
University of Glasgow
Glasgow, UK
frank.hopfgartner@glasgow.ac.uk
ABSTRACT campaign that allows researchers to benchmark news article rec-
Nowadays, most recommender systems provide recommendations ommendation algorithms in an offline [9] and an online [1, 5] set-
by either exploiting feedback given by similar users, referred to as ting. Given the content rich nature of news articles, as well as the
collaborative filtering, or by identifying items with similar proper- large numbers of users within NewsREEL who access news online
ties, referred to as content-based recommendation. Focusing on the [8], the lab can serve as a training ground to improve both content-
latter, this keynote presents various examples and case studies that based and collaborative filtering techniques.
illustrate both strengths and weaknesses of content-based recom-
mendation. 2. REFERENCES
[1] T. Brodt and F. Hopfgartner. Shedding light on a living lab:
1. OUTLINE the CLEF NewsREEL open recommendation platform. In
Proceedings of IIiX’14, pages 223–226, 2014.
Given the information overload that we are facing nowadays,
[2] F. Hopfgartner and J. M. Jose. Semantic user modelling for
tools and systems are required that help us to filter through large
personal news video retrieval. In Proceedings of MMM’10,
information and product spaces. Recommender systems approach
pages 336–346, 2010.
this task by predicting the preference or rating that a user would
[3] F. Hopfgartner and J. M. Jose. Semantic user profiling
give to an item. Two methods (or combinations thereof) to pro-
techniques for personalised multimedia recommendation.
vide these predictions dominate the field – collaborative filtering
Multimedia Syst., 16(4-5):255–274, 2010.
and content-based recommendation. Collaborative filtering meth-
ods exploit information on users’ behaviours or preferences to iden- [4] F. Hopfgartner and J. M. Jose. An experimental evaluation of
tify their interests and predict items that similar users showed inter- ontology-based user profiles. Multimedia Tools Appl.,
est in (e.g., [6, 7, 10, 11]). On the other hand, content-based recom- 73(2):1029–1051, 2014.
mender systems aim to identify users’ interests based on analysing [5] F. Hopfgartner, B. Kille, A. Lommatzsch, T. Plumbaum,
the actual content of items that they interacted with. T. Brodt, and T. Heintz. Benchmarking news
This talk first introduces the conceptual idea behind content- recommendations in a living lab. In Proceedings of
based recommendation. Representative systems and studies are CLEF’14, pages 250–267, 2014.
presented that illustrate the advantage of semantic metadata, as well [6] F. Hopfgartner, J. Urban, R. Villa, and J. M. Jose. Simulated
as the challenges that come with an automated analysis of content, testing of an adaptive multimedia information retrieval
especially in the multimedia domain. system. In Proceedings of CBMI’07, pages 328–335, 2007.
Following this overview, two methods to capture users’ interests [7] F. Hopfgartner, D. Vallet, M. Halvey, and J. M. Jose. Search
in items are introduced, namely explicit and implicit user profiling. trails using user feedback to improve video search. In
Explicit user profiles are created by asking users to rate items in Proceedings of Multimedia’08, pages 339–348, 2008.
a collection. Implicit user profiles are created by gathering user [8] B. Kille, F. Hopfgartner, T. Brodt, and T. Heintz. The plista
interest based on implicit relevance feedback such as viewing or dataset. In Proc. of NRS’13, pages 14–21. ACM, 10 2013.
clicking behaviour online. Examples and case studies [2, 3, 4, 12] [9] B. Kille, A. Lommatzsch, R. Turrin, A. Sereny, M. Larson,
are presented that illustrate the advantages and limitations of both T. Brodt, J. Seiler, and F. Hopfgartner. Stream-based
techniques. recommendations: Online and offline evaluation as a service.
The talk ends with an overview of NewsREEL1 , an evaluation In Proceedings of CLEF’15, pages 497–517, 2015.
1
[10] D. Vallet, F. Hopfgartner, and J. M. Jose. Use of implicit
http://clef-newsreel.org/ graph for recommending relevant videos: A simulated
evaluation. In Proc. of ECIR’08, pages 199–210, 2008.
[11] D. Vallet, F. Hopfgartner, J. M. Jose, and P. Castells. Effects
of usage-based feedback on video retrieval: A
simulation-based study. ACM Trans. Inf. Syst., 29(2):11,
2011.
[12] J. Yuan, F. Sivrikaya, F. Hopfgartner, A. Lommatzsch, and
M. Mu. Context-aware LDA: Balancing Relevance and
Diversity in TV Content Recommenders. In Proceedings of
CBRecSys 2015, September 20, 2015, Vienna, Austria. RecSysTV’15, 2015.
Copyright 2015 by the author(s).