=Paper=
{{Paper
|id=Vol-1892/overview
|storemode=property
|title=Workshop on Recommendation in Complex Scenarios (ComplexRec 2017)
|pdfUrl=https://ceur-ws.org/Vol-1892/overview.pdf
|volume=Vol-1892
|authors=Toine Bogers,Marijn Koolen,Bamshad Mobasher,Alan Said,Alexander Tuzhilin
|dblpUrl=https://dblp.org/rec/conf/recsys/BogersKMST17a
}}
==Workshop on Recommendation in Complex Scenarios (ComplexRec 2017)==
https://ceur-ws.org/Vol-1892/overview.pdf
Workshop on Recommendation in Complex Scenarios
(ComplexRec 2017)
Toine Bogers Marijn Koolen Bamshad Mobasher
Department of Communication & Huygens ING, Royal Netherlands School of Computing
Psychology Academy of Arts and Sciences DePaul University
Aalborg University Copenhagen Netherlands United States
Denmark marijn.koolen@huygens.knaw.nl mobasher@cs.depaul.edu
toine@hum.aau.dk
Alan Said Alexander Tuzhilin
University of Skövde Stern School of Business
Sweden New York University
alansaid@acm.org United States
atuzhili@stern.nyu.edu
ABSTRACT a variety of constraints on which recommendations are interesting
Recommendation algorithms for ratings prediction and item rank- to the user and when they are appropriate. However, relatively little
ing have steadily matured during the past decade. However, these research has been done on how to elicit rich information about
state-of-the-art algorithms are typically applied in relatively straight- these complex background needs or how to incorporate it into
forward scenarios. In reality, recommendation is often a more the recommendation process. Furthermore, while state-of-the-art
complex problem: it is usually just a single step in the user’s more algorithms typically work with user preferences aggregated at the
complex background need. These background needs can often place item level, real users may prefer some of an item’s features more
a variety of constraints on which recommendations are interesting than others or attach more weight in general to certain features.
to the user and when they are appropriate. However, relatively little Finally, providing accurate and appropriate recommendations in
research has been done on these complex recommendation scenar- such complex scenarios comes with a whole new set of evaluation
ios. The ComplexRec 2017 workshop addressed this by providing and validation challenges.
an interactive venue for discussing approaches to recommendation The current generation of recommender systems and algorithms
in complex scenarios that have no simple one-size-fits-all-solution. are good at addressing straightforward recommendation scenar-
ios, yet more complex scenarios as described above have been
KEYWORDS underserved. The ComplexRec 2017 workshop addressed this
by providing an interactive venue for discussing approaches to
Complex recommendation
recommendation in complex scenarios that have no simple one-
size-fits-all solution.
1 INTRODUCTION While ComplexRec 2017 was the first edition of this workshop,
Over the past decade, recommendation algorithms for ratings pre- in recent years other workshops have been organized on related
diction and item ranking have steadily matured, spurred on in part topics. Examples include the CARS (Context-aware Recommender
by the success of data mining competitions such as the Netflix Systems) workshop series (2009-2012) organized in conjunction
Prize, the 2011 Yahoo! Music KDD Cup, and the RecSys Challenges. with RecSys [1–4], the CARR (Context-aware Retrieval and Recom-
Matrix factorization and other latent factor models emerged from mendation) workshop series (2011-2015) organized in conjunction
these competitions as the state-of-the-art algorithms to apply in with IUI, WSDM, and ECIR [5, 7–9, 15], as well as the SCST (Support-
both existing and new domains. However, these state-of-the-art ing Complex Search Tasks) workshop series (2015, 2017) organized
algorithms are typically applied in relatively straightforward and in conjunction with ECIR and CHIIR [11, 12].
static scenarios: given information about a user’s past item pref-
erences in isolation, can we predict whether they will like a new
2 FORMAT & TOPICS
item or rank all unseen items based on predicted interests?
In reality, recommendation is often a more complex problem: ComplexRec was organized as an interactive, half-day workshop.
the evaluation of a list of recommended items never takes place in a The workshop started with a keynote presentation by Dietmar Jan-
vacuum, and it is often only a single step in the user’s more complex nach about his work on session-aware recommendation, where a
background task or need. These background needs can often place recommender system has to adapt its suggestions instantly to the as-
sumed short-term interests of each user, usually based on the user’s
ComplexRec 2017, Como, Italy. most recent interactions with the site or app. The keynote presenta-
2017. Copyright for the individual papers remains with the authors. Copying permitted tion was followed by a single paper session, for which short papers
for private and academic purposes. This volume is published and copyrighted by its
editors. Published on CEUR-WS, Volume 1892.. and position papers of 2-4 pages in length were solicited. Accepted
submissions received short 10-minute presentations with 5 minutes
ComplexRec 2017, August 31, 2017, Como, Italy. Bogers et al.
for discussion. Evaluation criteria for acceptance included novelty, from DBpedia and can be manually adjusted through a proprietarily
diversity, significance, quality of presentation, and the potential developed software tool.
for sparking interesting discussion at the workshop. All submitted Lofi and Tintarev [13] discuss a first step towards analogy-based
papers were reviewed by the Program Committee. The second half recommendation by benchmarking the semantics of perceived
of the workshop featured 3-4 breakout groups corresponding to the analogies. Their results show that current word embedding ap-
participant’s interests in addition to the topics of the contributed proaches are still not not suitable to sufficiently deal with deeper
papers. Afterwards, the breakout groups reported back for more analogy semantics.
discussion on what was learned. Finally, Wibowo, Siddharthan, Lin, and Masthoff [16] tackle
the complex problem of package recommendation where utility
of combinations of items must also be considered, such as travel
2.1 Topics of interest
or fashion. They introduce both a new data set for this domain
Relevant topics for the ComplexRec workshop included: and propose several extensions to the existing matrix factorization
• Task-based recommendation (Approaches that take the framework.
user’s background tasks and needs into account when gen-
erating recommendations)
4 WEBSITE & PROCEEDINGS
• Feature-driven recommendation (Techniques for elicit- The workshop material (list of accepted papers, invited talk, and
ing, capturing and integrating rich information about user the workshop schedule) can be found on the ComplexRec work-
preferences for specific product features) shop website at http://complexrec2017.aau.dk. A summary of the
• Constraint-based recommendation (Approaches that workshop will appear in SIGIR Forum to increase cross-disciplinary
successfully combine state-of-the-art recommendation al- awareness of recommender systems research.
gorithms with complex knowledge-based or constraint-
based optimization) REFERENCES
• Query-driven recommendation (Techniques for elicit- [1] Gediminas Adomavicius, Linas Baltrunas, Ernesto William de Luca, Tim Hussein,
and Alexander Tuzhilin. 2012. 4th Workshop on Context-aware Recommender
ing and incorporating rich information about the user’s Systems (CARS 2012). In Proceedings of RecSys ’12. ACM, New York, NY, USA,
recommendation need (e.g., need for accessibility, engage- 349–350.
[2] Gediminas Adomavicius, Linas Baltrunas, Tim Hussein, Francesco Ricci, and
ment, socio-cultural values, familiarity, etc.) in addition to Alexander Tuzhilin. 2011. 3rd Workshop on Context-aware Recommender Sys-
the standard user preference information) tems (CARS 2011). In Proceedings of RecSys ’11. ACM, New York, NY, USA,
• Context-aware recommendation (Methods for the ex- 379–380.
[3] Gediminas Adomavicius and Francesco Ricci. 2009. RecSys’09 Workshop 3:
traction and integration of complex contextual signals for Workshop on Context-aware Recommender Systems (CARS-2009). In Proceedings
recommendation) of RecSys ’09. ACM, New York, NY, USA, 423–424.
• Complex data sources (Approaches to dealing with com- [4] Gediminas Adomavicius, Alexander Tuzhilin, Shlomo Berkovsky, Ernesto W.
De Luca, and Alan Said. 2010. Context-awareness in Recommender Systems:
plex data sources and how to infer user preferences from Research Workshop and Movie Recommendation Challenge. In Proceedings of
these sources) RecSys ’10. ACM, New York, NY, USA, 385–386.
[5] Matthias Böhmer, Ernesto W. De Luca, Alan Said, and Jaime Teevan. 2013. 3rd
• Evaluation & validation (Approaches to the evaluation Workshop on Context-awareness in Retrieval and Recommendation. In Proceed-
and validation of recommendation in complex scenarios) ings of WSDM ’13. ACM, New York, NY, USA, 789–790.
[6] Pedro G. Campos, Nicolás Rodríguez-Artigot, and Iván Cantador.
[7] Ernesto William De Luca, Matthias Böhmer, Alan Said, and Ed Chi. 2012. 2nd
3 ACCEPTED PAPERS Workshop on Context-awareness in Retrieval and Recommendation: (CaRR
2012). In Proceedings of IUI ’12. ACM, New York, NY, USA, 409–412.
A total of 7 papers were submitted to the workshop, which were [8] Ernesto William De Luca, Alan Said, Matthias Böhmer, and Florian Michahelles.
all reviewed by a program committee of international experts in 2011. Workshop on Context-awareness in Retrieval and Recommendation. In
Proceedings of IUI ’11. ACM, New York, NY, USA, 471–472.
the field. Five of these papers were accepted for presentation at the [9] Ernesto W. De Luca, Alan Said, Fabio Crestani, and David Elsweiler. 2015. 5th
workshop, resulting in an acceptance rate of 71.4%. Workshop on Context-awareness in Retrieval and Recommendation. In Proceed-
ings of ECIR ’15. Springer, 830–833.
The accepted papers focused on a variety of complex recom- [10] Joaquin Delgado, Ravi Kalluri, Krishnaja Gutta, Arun Krishna, and Devon Turner.
mendation problems. Delgado, Kalluri, Gutta, Krishna, and Turner [11] Maria Gäde, Mark Michael Hall, Hugo C. Huurdeman, Jaap Kamps, Marijn
[10] discussed the complexity inherent in personalized voice search Koolen, Mette Skov, Elaine Toms, and David Walsh. 2015. First Workshop
on Supporting Complex Search Tasks. In Proceedings of the First International
for Internet TV, which requires the generation of fresh, domain- Workshop on Supporting Complex Search Tasks, co-located with ECIR 2015.
specific, relevant and contextual recommendations under a variety [12] Marijn Koolen, Jaap Kamps, Toine Bogers, Nicholas J. Belkin, Diane Kelly, and
of personal and general constraints. Emine Yilmaz. 2017. Current Research in Supporting Complex Search Tasks.
In Proceedings of the Second Workshop on Supporting Complex Search Tasks, co-
Piazza, Süßmuth, and Bodendorf [14] investigate the usefulness located with CHIIR 2017. 1–4.
of 3D body scans for fashion product recommendations. They ex- [13] Christoph Lofi and Nava Tintarev.
[14] Alexander Piazza, Jochen Süßmuth, and Freimut Bodendorf.
tracted a variety of different body measures from this complex data [15] Alan Said, Ernesto W. De Luca, D. Quercia, and Matthias Böhmer. 2014. 4th Work-
source and showed that it significantly improved the recommenda- shop on Context-awareness in Retrieval and Recommendation. In Proceedings of
tion performance. ECIR ’14. Springer, 802–805.
[16] Agung T. Wibowo, Advaith Siddharthan, Chenghua Lin, and Judith Masthoff.
Campos, Rodríguez-Artigot, and Cantador [6] describe the con-
struction and composition of a semi-automatically constructed
context taxonomy for extracting context data from user reviews for
recommendation. The taxonomy is composed of semantic entities