Reducing seed noise in personalized PageRank

Abstract

Network based recommendation systems leverage the topology of the underlying graph and the current user context to rank objects in the database. Random-walk based techniques, such as PageRank, encode the structure of the graph in the form of a transition matrix of a stochastic process from which the significances of the nodes in the graph are inferred. Personalized PageRank (PPR) techniques complement this with a seed node set which serves as the personalization context. In this paper, we note (and experimentally show) that PPR algorithms that do not differentiate among the seed nodes may not properly rank nodes in situations where the seed set is incomplete and/or noisy. To tackle this problem, we propose alternative robust personalized PageRank (RPR) strategies, which are insensitive to noise in the set of seed nodes and in which the rankings are not overly biased towards the seed nodes. In particular, we show that novel teleportation discounting and seed-set maximal PPR techniques help eliminate harmful bias of individual seed nodes and provide effective seed differentiation to lead to more accurate rankings. We also show that the proposed techniques lead to efficient implementations, where existing approximation algorithms and/or parallel implementations for computing the PPR scores can be easily leveraged. Moreover, the proposed formulations are reuse-promoting in the sense that, it is possible to divide the work relative to individual seed nodes and cache the intermediary results obtained during the computation, and especially in systems with large query throughputs, it may be possible to cluster queries based on the partial overlaps between the seed sets and reduce the overall robust PPR computation costs. Experiment results show that the proposed techniques are efficient and highly effective in improving recommendations and eliminating unwanted bias due to imperfections in the seed set.

Like-tasted user groups to predict ratings in recommender systems

“…The MovieLens-100K dataset is widely used to assess recommender systems efficiency [4,11]. It was collected as a part of the GroupLens research project, through the MovieLens website.…”