In this paper, we describe our on-going research on the problem of predicting needed hyperlinks between pairs of Wikipedia pages $(u,v)$ that are not connected, yet show readers’ searches from $u$ to $v$. We propose a solution that first estimates how long will these searches last and then predicts new hyperlinks according to descending order of duration. Our initial experimental results show that our best solution achieves an AUROC of 0.77 on the Wikipedia Clickstream dataset and a precision@20% of 1.0 and significantly beats the baselines.