Duration: 1:01:25

90 views

1

Social media features such as social explanations of recommendations (e.g., "Amit liked this") and feeds of activity around items (e.g., listening to songs, visiting URLs, reading books) mean that we are often exposed to our friends' preferences online. This raises the question of how those preferences interact with our own, both at the individual level – how do people influence each other's choices – and globally – how do these preferences affect the diffusion of items through networks? In this talk, I will describe a set of micro-experiments and large-scale observational studies that shed some light on this question. First, I will demonstrate the existence of preference locality – friends having similar item preferences – in social networks that makes recommendations computed from friends' activities comparable in accuracy to those computed from all users' data. To understand how such locality emerges, I will describe two studies: one investigating the influence of social explanations using a randomized experiment, and another studying the influence of feeds on users' actions using activity data from multiple social networks. Both studies conclude that social influence plays only a secondary role: models of people's decision making indicate that a majority of people's actions can be explained by a process where they follow their own preference for items. These results point toward the value of accounting for individual preference when modeling diffusion and suggest ways for incorporating social influence in recommendation algorithms.