The Lifecycles of Apps in a Social Ecosystem
http://arxiv.org/pdf/1503.06870v1.pdf
Two main concepts introduced in this paper are sociality and popularity. Popularity of an application is just defined the raw number of users of the application. The sociality of an application is how likely a user will adopt the app given friends who already use the app.
The paper poses several interesting questions which relate to topics we discussed recently in class, namely: cascading behavior in networks, how strong and weak ties in a network can predict adaptation of a behavior. The paper attempts to answer this questions within multiple frameworks discussed in class.
One such question posed is, assuming a non-user A and a user B who is A’s friend, does the probability of A adopting the application depend on how similar user B is to the median user of the application or on how similar user B is to user A? This comparison attempts to measure the effectiveness of network cascades against the idea of strong ties when it comes to behavior adaptation.
Another interesting question posed was: say A has multiple friends using the application. Does A’s probability of adopting the application increase if his/her friends who are users have edges between them as opposed to no edges? This question attempts to measure the effectiveness of having connected friends against having disconnected friends who use the app. Basically, how does having a local neighborhood of friends using the app compare to not having a local neighborhood but still having similar number of friends using the app.
A third interesting question posed was: can we use an app’s social, demographic, retention, and temporal features to predict whether or not it will be successful in the long term?
For this question a binary classifier was trained. Using different set of features in the experiment, multiple random forest models were trained with the same dataset. The first model used temporal features which yielded best results, at 70% accuracy. Interestingly, the most important features were median number of users in months 8,9,12. High weekly minimum number of users was also a indication of stability. The set of features with second highest success were individual user attributes such as the fraction of app users that were also active Facebook users for the past k out of 7 days. High sociality, turns out is a negative indicator of success.
Seeing multiple network frameworks being used in order to explain application adaptation in a Facebook friends network gives interesting insights on the power of multiple network effects.