Can we tell if a user will rate a certain photo higher than average, or predict which photos in general that person will rate above average?This formulates the question of whether a better prediction of the rating behavior of a user can be made using the previously stated inputs, thereby improving the matching system of paiq.This research therefore strives to evaluate and apply machine learning techniques in novel ways, in order to obtain accurate predictions of a user’s scoring of photos he/she has never actually seen.
The University of Twente spin-off paiq (founded in 2005) is not a traditional dating site.
Behind the scenes, neural networks are used to match users in their quest for a suitable partner, mainly using ‘matching questions’ about their lifestyle and personality.
Paiq is constantly looking to improve their matching system.
Their website has a web application in which users can rate photos of possible matches.
Enough users use this option so that their database contains tens of millions of rankings.
Paiq wants to include appearance in their matching system.While writing a naive algorithm to calculate the ranking of a photo is the base line of this research, this final project focuses on more complicated situations.The inputs for the algorithm contain the average ratings of photos, created from photo ratings done by users, and the past photo rating voting behavior of a user.New Dutch dating site Paiq matches people with artificial neural networks, a form of artificial intelligence trying to imitate the way human brains work.People who register first have to enter personal characteristics and a picture.Paiq then matches people based on intelligent rules, which in their turn our based on psychological insights.