Written by: Shailendra Rao, Tom Hurlbutt, Clifford Nass, and Nundu JanakiRam
Comments: Shaun
Summary:
This paper was about recommender systems and how to make them better in a dating site situation. It starts out by saying the two things it investigates is 1) adding personal photos to increase a person self-awareness when they are creating a profile, and 2) providing recommendations based on personal question answers. It also did tests based on showing a person results throughout the question answering situation or showing a user all of the results at the end. Recommender systems are known for creating a user profile of a person and recommending things to that user based on the information that is has received. Sometimes the recommender systems do a very poor job or recommending people or items to someone, and that can prove to be very frustrating for the user. This experiment used personal photos as a means to get the user to have more "self-awareness" so that they would be consistent in the questions that they answer. The experiment consisted of a 2x2 web-based study with a dating algorithm called "MetaMatch". The users were given a set of questions, but for this experiment no matter how the user answered they were going to get the same set of users recommended to them. Half of the participants had their photo shown on the left side of the site while they answered questions, and the other half did not. The participants in this experiment were told that the answers to the questions they were to be given would be used in the algorithm that would return a profile match to them. This was not the case because they were actually given the same set of profiles with the only difference being gender specific to the participants taste. There was a total of 40 questions for the profile, and half of the users would receive a match after each block of 10 questions, while the other would receive results at the end of the 40 questions. They found that having ones self portrait shown while answering questions did have an impact on the reaction of the user to the matches they received. They measured change of response strategy, if a user liked the recommendations, and what frustration level the users were at based on the matches they received. Results are shown below.
The results show the half that was shown their photo while answering questions, and the half that was not shown a photo. Those halves were then separated into those who were shown photos throughout the blocks of 10 questions, and those who were given all of the results at the end. Those who were shown their photos were less likely to change their response strategy because they found it difficult to do so. The self-reflection "superseded" their ability to adapt to the truly bad recommendations of the dating algorithm. Those who did not see their photo were almost motivated to change their answers so that they would start receiving better matches.
Personal photos can only work for certain recommender systems because things like netflix aren't allowed to ask for your photo. Those systems that can use a photo have a better chance of getting consistent answers from a user, and it will become easier for that system to recommend good things to the user. A lot of dating sites don't allow user feedback right away when given a match so this hinders it's ability to test if it has done a good job or not. If someone gets a really bad match they would prefer to mention it to the system so that it can adapt and begin giving them better matches.
Discussion:
I thought this was interesting because I've tried creating a recommender system in one of my classes and never realized how difficult it was to get user feedback and information so that you can create a profile for them. That user profile is crucial because one mess up there and you may frustrate your user into not using your system anymore. I don't think there were any flaws in this because all of the users were told the same things, and matched up with the same profiles so that they feedback at the end of the experiment would be consistent. I think future work could include making dating websites less stressful to use because they would be getting better results by simply showing a user their photo while answering all of the questions provided by the website. There also needs to be something that doesn't involve your photo so recommender systems such as movie or music recommender could also become better at making suggestions to a user.
No comments:
Post a Comment