Several works have outlined the fact that the mobility in intermittently connected wireless networks is strongly governed by human behaviors as they are basically human-centered. It has been shown that the users' moves can be correlated and that the social ties shared by the users highly impact their mobility patterns and hence the network structure. Tracking these correlations and measuring the strength of social ties have led us to propose an efficient distributed tensor-based link prediction technique. In fact, we are convinced that the feedback provided by such a prediction mechanism can enhance communication protocols such as opportunistic routing protocols. In this paper, we aim to bring out that measuring the stabilities of the link and the proximity at two hops can improve the efficiency of the proposed link prediction technique. To quantify these two parameters, we propose an entropy estimator in order to measure the two stability aspects over successive time periods. Then, we join these entropy estimations to the tensor-based link prediction framework by designing new prediction metrics. To assess the contribution of these entropy estimations in the enhancement of tensor-based link prediction efficiency, we perform prediction on two real traces. Our simulation results show that by exploiting the information corresponding to the link stability and/or to the proximity stability, the performance of the tensor-based link prediction technique is improved. Moreover, the results attest that our proposal's ability to outperform other well-known prediction metrics.