Query-oriented Graph Clustering

There are many tasks including diversified ranking and social circle discovery focusing on the relationship between data as well as the relevance to the query. These applications are actually related to query-oriented clustering. In this paper, we firstly formulate the problem, query-oriented clustering, in a general form and propose the two measures, query-oriented normalized cut (QNCut) and cluster balance to evaluate the results for query-oriented clustering. We develop a model, query-oriented graph clustering (QGC), that combines QNCut and the balance constraint based on cluster balance in a quadratic form. In the experiments, we show that QGC achieves promising results on improvement in query-oriented clustering and social circle discovery.