“My Dating Site Thinks I’m a Loser”: Effects of Personal Photos and Presentation Intervals on Perceptions of Recommender Systems

Authors

Shailendra Rao
Stanford University
shailo@stanford.edu


Tom Hurlbutt
Stanford University/ Intuit
hurlbutt@cs.stanford.edu


Clifford Nass
Stanford University
nass@stanford.edu


Nundu JanakiRam
Stanford University/ Google
nundu@google.com


Summary


This study explores the effects that bad recommendations from recommender systems can have on the individuals that use those systems.  For example, if someone's TiVo mistakenly thinks that they are gay and recommends television programs that would appeal to those that are gay, then the user might erroneously watch more "guy" programming to compensate for their miscategorization.

The situation that is explored in this paper is online dating.  People will lie about themselves, omit information about themselves, or submit misleading pictures of themselves to present their "ideal" self.  Giving false information like this only skews the system and matches users with people who differ substantially from what they were expecting.  This leads to frustration for both parties involved and now the dating site cannot effectively carry out its purpose or end goal.

The study used a new algorithm called MetaMatch, which took the answers to a set of personal questions and used them to purposefully suggest undesirable matches.  After receiving the matches, the participants rated their recommendations and answered a post questionnaire.

During the questionnaire, half of the participants had their submitted picture posted next to the questions as they answered them and the other half had no picture.  Also, half of the participants were shown 4 possible (undesirable) matches  after each set of 10 questions and the other half were only shown matches after they finished the entire questionnaire.

The results of the post questionnaire are represented in the graphs above.  The study found that displaying a personal photo can have a stabilizing effect amongst participants. Having an element of self-reflection during the questionnaire tended to make the participants not change their answers to the questions to compensate for their bad matches. Also, intermediate recommendations tended to make the user change how much they approved of the bad match at the end of the questions.

Discussion
I thought this study was one of those studies that proves something that you already suspect.  It would be interesting to see what online dating companies do with this information or with more indepth studies.  I do feel like only testing 56 people was a bit small for the user study though. I would like to have seen a more exhaustive study that included thousands of participants.