The conference has almost 1000 submissions (so huge number of submissions...), and both full and short tracks have around 23% acceptance rate.
The word cloud reflects hot keywords in the accepted papers, which includes deep learning and search etc.
Keynotes:
There were three keynote speakers from big companies such as MS, Google.
1. Toward Data-Driven Education
Rakesh Agrawal (Data Insights Laboratories)
Rakesh Agrawal (Data Insights Laboratories)
2. Personalized Search: Potential and Pitfalls
Susan Dumais (Microsoft Research)
Susan Dumais (Microsoft Research)
3. A Personal Perspective and Retrospective on Web Search Technology
Andrei Broder (Google Research)
Andrei Broder (Google Research)
The second keynote was interesting for me as the keynote speaker talked about personalization in the context of search, and mentioned the User Modeling(UM) aspect.
Susan talked two types of UM (or user profiles), one is local profile which can be stored in PC, and the other one is cloud profile. The local profile is good considering user privacy as the profile is in the local PC for personalization, i.e., the ranking results will be personalized based on the local profile of a user. However, it suffers from in efficiency, e.g., due to the lack of portability, it will be hard to reuse it if the user changes the working environment (e.g., change PC). Personal score was carried out by content matching as well as features based on interaction history.
She also talked about evaluation of personalization alternatives, offline and online ones. As we can expect, the former one is safe to exploit many different alternatives. On the other hand, the later one (e.g., A/B testing) has more accurate evaluation, and with some challenges. Explicit feedback from users (e.g., asking users about the personalization is good or not) can be a good indicator, however, it also might change the user search behavior. On the other hand, implicit feedback could be noisy.
Another interesting point is personalization can also provide interesting items (serendipity)...
Tutorials
There were eight tutorials and I chose the tutorial: "Data-Driven Behavioral Analytics: Observations, Representations and Models", which was given by Dr. Meng Jiang and Dr. Jiawei Han
The tutorial is about human behavior analytics, which is one of the six disruptive research areas defined by Department of Defense. They introduced many models incorporating different factors of social network information into traditional #RecSys approaches such as Matrix Factorization (MF).
The tutorial is about human behavior analytics, which is one of the six disruptive research areas defined by Department of Defense. They introduced many models incorporating different factors of social network information into traditional #RecSys approaches such as Matrix Factorization (MF).
Sessions
I also attended #RecSys session. As this conference is about IR, there were many models introduced by different problems, and MF seems like the dominated one. Surprisingly, none of the first authors came to present those papers.
Maybe due to the venue?, there were many people didn't come either for presenting or taking CIKM cup awards...:). It was a good experience to attend the first IR conference for me, and the next CIKM(2017) will be at Singapore.