Learning collective behavior of social media sites using Variant of k-means algorithm In this paper we have implemented a new k-means variant for clustering. Huge amount of data is generated on social media websites which challenges us to predict collective behavior. Collective behavior means to understand how the individual behaves in social networking environment. There is problem of prediction on sites as there are many numbers of actors. So because of this problem a new technique of edge centric clustering is carried out here which extracts sparse social dimensions. A k-means variant algorithm is then implemented for clustering which reduces the time required for clustering. The experimental results shows that variant of k-means has given better results than the other k-means algorithm. We could see this by the comparison shown of two k means algorithms.