Cross-domain recommender systems and transfer learning approaches are useful to help integrate knowledge from di erent places, so that we alleviate some existing problems (such as the cold-start problem), or improve the quality of recommender systems. With the advantages of these techniques, we host the rst international workshop on intelligent recommender systems by knowledge transfer and learning (RecSysKTL) to provide such a forum for academia researchers and application developers from around the world to present their work and discuss exciting research ideas or outcomes. The workshop is held in conjunction with the ACM Conference on Recommender Systems 2017 on August 27th at Como, Italy.
• Information systems → Recommender systems;
Recommender systems, as one of well-known Web intelligence applications, aim to alleviate the information overload problem and produce item suggestions tailored to user preferences. Typically, user preferences or tastes are collected through usersŠ implicit or explicit feedback in various formats, such as user ratings, online behaviors, text reviews, etc. Also, user feedback on di erent items can be collected from several systems or domains. The diversity of feedback formats and domains provides multiple views to usersŠ preferences, and thus, can be helpful in recommending more related items to users. Cross-domain recommender systems and transfer learning approaches propose to take advantage of such diversity of
haviors on social networks to predict their preferences on movies (e.g., Net ix, Youtube) or music (e.g., Pandora, Spotify).
taste on electronics by using his or her preference history on books based on the data collected from Amazon.com. From one context to another: We may collect a userŠs preferences on the items over di erent time segment (e.g., weekend or weekday) and predict her preferences on movie watching within another context (e.g., companion and location). From one task to another: It may be useful for us to predict how a user will select hotels for his or her vocations by learning from how he or she books the tickets for transportation.
us to infer social connections by learning from the structure of heterogeneous information network. 2
This workshop intends to create a medium to generate more practical and e cient predictive models or recommendation approaches by leveraging user feedbacks or preferences from multiple domains.
This workshop will be bene cial for both researchers in academia and data scientists in industry to explore and discuss di erent de nition of domains, interesting applications, novel predictive models or recommendation approaches to serve the knowledge transfer and learning from one domain to another.
As this is the rst time to host such a workshop on knowledge transfer and learning in the area of recommender systems, we also expect to collect feedbacks from the workshop, so that we can better organize the workshop in the future. 3
The topics of interest include (but are not limited to):
Cross-domain recommendation Context-aware or time-aware recommendation Multi-criteria recommender systems
Novel applications
Knowledge transfer for content-based ltering Knowledge transfer in user- and item-based collaborative ltering Transfer learning of model-based approaches to collaborative ltering
Deep Learning methods for knowledge transfer
Addressing user feedback heterogeneity from multiple domains (e.g. implicit vs. explicit, binary vs. ratings, etc.) Multi-domain and multi-task knowledge representation and learning Detecting and avoiding useless knowledge transfer Ranking and selection of auxiliary sources of knowledge to transfer from Performance and scalability of knowledge transfer approaches for recommendation
Beyond accuracy: novelty, diversity, and serendipity of recommendations supported by the transfer of knowledge Performance of knowledge transfer systems in cold-start scenarios Impact of the size and quality of transferred data on target recommendations Analysis of the amount of domain overlap on recommendation performance
We thank the contributions by the following workshop program committee members:
Alejandro Bellogín, Universidad Autónoma de Madrid Steve Bourke, Schibsted Media Group Iván Cantador, Universidad Autónoma de Madrid Liang Dong, Google, Inc.
Mehdi Elahi, Free University of Bozen Negar Hariri, Apple, Inc.
Mahesh Joshi, Linkedin Bin Li, Data61, Australia Zhongqi Lu, Hong Kong University of Science and Technology, Hong Kong, China Cataldo Musto, University of Bari "Aldo Moro" Denis Parra, Ponti cia Universidad Catolica de Chile Alan Said, University of Skövde, Sweden Yue Shi, Facebook, USA Fatemeh Vahedian, DePaul University Saúl Vargas, Mendeley Tong Yu, Carnegie Mellon University, USA Fuzhen Zhuang, Chinese Academy of Sciences, China Yong Zhuang, Carnegie Mellon University, USA 5
This year, we received 11 valid submissions. We nally accepted 5 long and 2 short papers. The copyrights of the accepted papers are held by the owner/author(s). All of the accepted papers will be published by CEUR workshop proceedings1. Hopefully, we can collaborative with some journals, and invite the extended version of the accepted papers to a special issue in the journal. 6
We thank the workshop organizing committee for giving us the opportunity to host this workshop in conjunction with ACM RecSys 2017. And we appreciate the contributions by the authors and program committee members.