Peer recommendation by using pattern mining to generate candidate keywords in attributed graphs

We introduce a query-based community retrieval algorithm which works on an attributed graph. On issuing a query with a vertex and core value, our model first generates relevant candidate keyword sets based on the attributes of nodes. The rapid keyword extraction algorithm is used to generate relevant candidate keyword sets. For each relevant keyword set, it recommends possible research peers for the queried author. In order to facilitate efficient community search in the graph, we used the Core-Label Tree indexing which provides an efficient way to model a dynamic network. The recommended communities hold peer attributes which are maximally congruent with keyword sets of another peer. We also proposed three different models to structure the connectivity among the nodes and compared their results. We also evaluated the built models. Our results show that the proposed models are efficient to predict relevant communities.