19 Maggio 2015
San Francesco - Sagrestia
Copying, sharing and linking have always been important for the functioning and the growth of the World Wide Web. Two recent copying trends which have emerged are social content curation, and social logins. Social curation involves the copying, categorization and sharing of links and images from third party websites on the social curation website. Social logins enable the copying of user identities and their friends from an established social network such as Facebook or Twitter, onto third party websites. In this talk, we chronicle our ongoing work on Pinterest, a popular image sharing website and social network. The highly active user community on Pinterest has been instrumental in making social curation a mainstream phenomenon. Interestingly, a large fraction (nearly 60%) of the users have also linked their Pinterest accounts with Facebook and have copied their Facebook friends over onto the new website. Thus, using a large dataset crawled from Pinterest, we uncover both the practices used for sharing content, as well as how the copying of friends has helped the content sharing. We find that social curation tends to copy and share hard-to-find niche interest content from websites with a low Alexa Rank or Google Page Rank. And users tend to agree with each other on the curation categories. Using user preference and image-related features drawn from a state-of-the-art deep convolutional network, we develop a prediction cascade to automate a large fraction of social curation actions. On the other hand, Pinterest users can also copy friends from Facebook, or Twitter. We find that this copying of friends create a community with higher levels of social interaction; thus social logins serve as a social bootstrapping tool. But beyond bootstrapping, we also find a weaning process, where active and influential users tend to form more links natively on Pinterest interact with native friends rather than copied friends.
Zhong , Changtao - King's College - London