A Scalable Method for One-Mode Projection of Bipartite Networks Based on Hadoop Platform

People look for models and methods to organize, classify, compress and filter the information due to the difficulty in maintenance and using immense sources of information. The bipartite graphs are particularly useful among the variety of presenting methods such as recommender systems. Most of the bipartite networks tend to cluster one side of graph behavior to recognize communications and interactions between members of that side and discover similar members. The one-mode projection technique is widely used for this purpose. However, parts of the primary information of the original bipartite graph is missed under the projection. So we need to exploit a method for determining the weights that yield projected edges in a way that minimizes information loss. While such methods exist, the majority of investigated databases in the field of bipartite network projection are huge, consequently, executing a projection procedure takes lots of times. In this paper, we propose a scalable method based on resource allocation for bipartite network projection. It provides a high performance while preserving precision through transferring the needed operations on a distributed platform like Hadoop. Moreover, as a case study, we evaluate the performance of the presented scalable algorithm in the field of social network which results in short projection operation time in comparison to the undistributed mode. Also, we compared our proposed method with a collaborative filtering method, a well-known algorithm in the recommendation field and as a result, our method had higher overall execution speed. With using the largest dataset of our experiments, the Orkut dataset, the proposed method has higher speed than the scalable CF by 33 %. Then, we evaluate the scalability of the introduced method by a scalability metric namely Speedup, which showed good scalability.