CS Forum: John O'Donovan, University of California

CS department's public guest lecture on 'Interaction Design and Evaluation for Recommender Systems'. Lecture is open to everyone free-of-charge.

Speaker: John O'Donovan
Speaker affiliation: University of California, Santa Barbara
Host: Antti Kangasrääsiö
Time: 11:15 (coffee at 11:00)
Venue: T2, CS building

Interaction Design and Evaluation for Recommender Systems

Abstract

Search and recommendation algorithms have become standard tools that most of us interact with in our daily lives.  Whether it is a movie from Netflix, an advert on Facebook, a result from a Google query or a product recommendation from Amazon, these complex algorithms work behind the scenes to passively or actively learn about user tastes and preferences. This information is used in turn to personally tailor the information space for the user.  However, these systems are not always accurate, and many of us have probably asked questions akin to "why does Netflix keep recommending sad movies to me?", or "why is Amazon trying to sell me a guitar?".  The problem, originally highlighted by Herlocker et. al., involves stale or incorrect data, algorithm transparency and result explanation.  If the mechanics and reasoning of a search or recommendation algorithm can be communicated to a user in the right way, it can improve acceptance of the prediction and trust in the system as a whole.  Better still, by mapping the mechanics of information filtering algorithms into visual spaces, we support the notion of user control over the algorithm at recommendation time, helping to build trust, improve user experience and alleviate issues arising from stale or otherwise incorrect user profile information. This talk will cover current research at UCSB's Four Eyes Lab that focuses on these questions.  I will introduce the key concepts and challenges of interacting with such artificial advice givers, and introduce a novel system known as MoodPlay, that recommends music based on an interactive latent space of mood information from a database of music artists.   

Speaker Bio

John O'Donovan is an associate research scientist and principal investigator at the Computer Science Department, University of California, Santa Barbara.  John received his PhD in Computer Science in 2008 from University College Dublin, Ireland, advised by Prof. Barry Smyth.  His research background is in AI, with a focus on recommender systems. He has a particular interest in modeling trust in social networks, from the perspective of mining big network data, and also from the HCI perspective.  Dr. O’Donovan has published more than 50 research papers in peer reviewed conferences and journals. His research on human computer interaction and intelligent interfaces has won multiple best paper awards, including at SocialCom 2013 and IEEE CogSima 2014 conferences.  John has served on program committees and as reviewer for more than 20 conferences and journals, including ACM RecSys, ACM IUI, ACM CHI, WWW, KDD, IJHCS, ACM TOIT and ACM TIST. He served as general co-chair of the 2016 ACM international conference on Intelligent User Interfaces, and as an associate editor for ACM TiiS journal special issue on Human Interaction with Artificial Advice Givers. He is the recipient of the inaugural award for most influential research paper in the ACM IUI conference series.

**

The speaker is visiting the PML research group and will be available for lunch and discussion after the talk until 14 o'clock.