Throughout my research years, I have always tried to grasp why & how users behave so unpredictably, whilst seeking to improve their experience.

My research has mainly been in the field of recommender systems. I have investigated some key factors which might lead users to accept an item recommended by the system, before eventually deciding to adopt it on a more regular basis. One particular focus was on diversity among recommended items. For the first time, Dr. S. Castagnos and I were able to show that users need some diversity (and not just accuracy) in order to finalise their choice before a purchase.

Another research theme has been to re-think users’ interactions to improve algorithms. Today’s popular five-star rating scales can create up to 40% of data noise or inconsistencies. We proposed a simple comparisons modality. With multiple users-studies, we showed that our approach can be 20% more stable, and highly appreciated by users.


Comparisons: I prefer A to B

Users are often asked to express a rating on a five-star scale. Unfortunately, ratings have been shown to introduce up to 40% of inconsistencies or noise. This is a serious issue: such data serves to create preference-models, which are in turn used to compute the recommendations. We proposed to replace ratings with a simple comparison modality (“I prefer A to B”). With multiple users-studies, we showed that our approach can be 20% more stable, and that users are in favour of relying on such an interaction.

Diversity among recommended items

Nearly all of the recommender research has focused on making algorithms which provide the most accurate suggestions. However, there is much more to users’ satisfaction than pure accuracy. I investigated diversity. For the first time, Dr. S. Castagnos and I were able to show that users actually need some diversity in order to make a confident purchase decision. We proposed a novel time-dependant diversity model.

Layout vs. content

I had the opportunity to analyse the respective effects of layout and content throughout my research. In the case of dynamic-critiquing for instance, Dr. J. Zhang and I created a collection of meaning-augmented icons, to replace textual critiques. The results were that users used them more frequently and felt it required less effort.

User acceptance, and possible adoption

Understanding what leads users to accepting (clicking on) recommended items, and later deciding to adopt this system (rather than another) has been, and remains one of my key interests. In order to decipher user’s motivation, I have designed and carried out over ten user studies, gathering over 600 participants’ feedback.

Intelligent systems

The user modelling community has long sought to make intelligent systems. The emergence of the Web has meant that users’ traces have become abundant (through log files for instance). I had the chance to work on users’ explicitly and implicitly expressed preferences, and at times I used an eye-tracker to establish links between users’ action and actual associated meaning.


Feeling adventurous, wanting to know more details? Then delve into the publications & posters without second thought.

Supervised Projects

Throughout my Ph.D. and post-doc, I had the pleasure of supervising the projects of the following excellent students:

  • Emeline Schmidt and Magali Kamalski. Developpment of a movie comparison framework, and running of three users-studies. (July-December 2010). Upcoming papers.
  • Thomas Girard. TagMatrix – a Data Visualization for TraceTrack. Improvements on the Lama Framework. (June 2009) [pdf]
  • Aurelia Rochat. Tomorrow’s music player: a spinning interface. (January 2009) [pdf]
  • Lucas Maystre: TraceTrack a Recommender System Interface That Displays Your Listening Habits. (January 2009) [pdf]
  • Ganesh Venkateshwaran. Mobile Critiquing Interface. (January 2008). [co-supervised with J. Zhang]
  • Joël Schintgen. Implementation of an Interactive travel Map. (January 2008) [co-supervised with L. Chen] [pdf]


I enjoy discussing an presenting my work, and had the opportunity of talking at the following events.

  • Eye-Tracking Product Recommenders’ Usage. At the ACM International Conference on Recommender Systems, Barcelona, Spain (September 2010).
  • User Perceived Qualities and Acceptance of Recommender Systems: the Role of Diversity. Presentation of my thesis work. EPFL (April 2010) and LORIA, Nancy (July 2010).
  • Découvrir les interactions Homme-Machine. Open day for high school students, EPFL (March 2010) and Open day doctoral school, EPFL (March 2010).
  • Explicit and Implicit User Preferences, a Human Computer Interaction Perspective. At the Workshop on Semantic User Descriptions and their Influence on 3D graphics and VR, as part of the SEMINAIRE DU TROISIEME CYCLE ROMAND, VRLAB – EPFL (November 13, 2008)
  • User Technology Adoption Issues in Recommender Systems. At Networking and Electronic Commerce Research Conference, Lake Garda, Italy (October 18-21, 2007).


I take interest in reviewing papers and have been a Program Committee Member for the IUI09 conference.

During my Ph.D. I was a teaching assistant for my professor’s Human-Computer Interaction master course. I contributed to the elaboration of exercises and exams, guidance of student projects, and grading of the work. I also taught the course during two months (medical leave of my Prof.).