Recommendation Systems are an important tool for aiding discovery of content such as movies, books, and music. Generating personalised recommendations that deliver serendipitous suggestions to the user is a key factor in determining user satisfaction. In this paper, we propose a recommendation algorithm that uses a community-structure based link prediction approach. The proposed link prediction algorithm predicts the links that are most likely to be formed in the network based on two new proposed metrics. These metrics take into account the effects of both the network structure and the attribute information of the items for predicting probable links. The links predicted can be presented as recommendations to the user. An item network consisting of item associations derived from the attribute information usually displays a community structure with items of similar nature being clustered together. Our approach utilises this community structure to expose implicit links in the item network and provides non-obvious recommendations to the user. © 2017 IEEE.