Methodological Analysis of Personalization in Urban Recommender Systems by Distance Measures

Citizens increasingly rely on urban recommender systems (URS's) to plan daily activities in the physical urban world. Nonetheless, there is a growing concern that personalization in URS's enforce urban filter bubbles. Existing research on this topic is still limited, especially methodologically. One problem is the use of a limited set of distance measures for analyzing differences between item sets returned by URS's under different conditions. Another problem is the lack of advanced analysis models for investigating and comparing the relative impact of different conditions on returned item sets. In this paper, we explore the use of different set distance and geospatial distance measures and multivariate distance matrix regression (MDMR) to assess the relative impact of different determinants of item sets. The analysis of data collected from Google Maps yielded more nuanced conclusions about filter bubbles and personalization when geospatial distance measures were used. Also, search language rather than search location is found to dominantly predict which items URS's return.