Link prediction is an active area within the broader research on social network analysis (SNA) that tries to predict future links using a social network structure. The purpose of link prediction is to use the current structure of the social network to predict interaction between the nodes that will be generated in the future. A mathematical formulation of link prediction methods can be articulated as: “Given a pair of nodes u and v in the current social network, how likely is it that u will interact with v in the future?”

This paper is an interesting work that uses a distributed learning automata–based approach in conjunction with fuzzy concepts for link prediction. While distributed learning automata and fuzzy set theory have both been studied in detail earlier, this is the first work that combines these two concepts for the purpose of link prediction.

Additional strengths of the paper are the detailed preliminary information and description of the problem space and the commonly used similarity measures. This establishes the context for the reader in great detail, and exposes new students to this field of study in a gentle manner. Another strength is the detailed empirical results that use five different datasets and compare the proposed algorithm to 10-plus previously known algorithms.

A concern regarding the proposed algorithm is that the running time is almost two orders of magnitude higher than the other algorithms used in the comparative analysis. This then raises the secondary concern that it may be possible to reconfigure the other algorithms to trade execution time for higher prediction accuracy. Therefore, it remains unclear how the proposed algorithm would compete against the reconfigured versions of algorithms used in comparative analysis.

An obvious idea for a future extension of this work is to compare the proposed model with an ensemble model that simply combines the results of the three best previously known algorithms. Considering the fact that the proposed algorithm takes a significantly longer time, it is conceivable that three prior known algorithms and the ensemble model together take less time than the proposed approach and deliver higher quality results.