An Online Algorithm for Smart Broadcasting in Social Networks
Ali Zarezade, Utkarsh Upadhyay, Hamid R. Rabiee, Manuel Gomez-Rodriguez
Everybody is a broadcaster
Not all posting times are the same.
\( r(t) \): position of the highest ranked post by our broadcaster on the follower's feed.
Other broadcasters can be bursty with time varying rates, which we cannot control.
We would like to control our rate of posting \( u(t) \) to minimize \( r(t) \).
For a very general class of behavior of other broadcasters, the optimal rate of posting is \( u^{*}(t) = c \times r(t) \).
Since \( u(t) = 0 \), so we do not need to post at all.
We take one sample from \( exp(c) \) to determine our next post's time.
If other broadcasters post in the meanwhile, we may revise our planned posting time.
However, we may not update our time if we planned to post was earlier than the new sampled time.
After we post, we start this process all over again.
\( 72\% \) average drop in rank, lowered for all users
Time at top \( \int_{0}^{t} \mathbb{I}(r(\tau) \lt 1)\, d\tau \)
Questions?