Local is good: A fast citation recommendation approach

Abstract

Finding relevant research works from the large number of published articles has become a nontrivial problem. In this paper, we consider the problem of citation recommendation where the query is a set of seed papers. Collaborative filtering and PaperRank are classical approaches for this task. Previous work has shown PaperRank achieves better recommendation in experiments. However, the running time of PaperRank typically depends on the size of input graph and thus tends to be expensive. Here we explore LocRank, a local ranking method on the subgraph induced by the ego network of the vertices in the query. We experimentally demonstrate that LocRank is as effective as PaperRank while being 15x faster than PaperRank and 6x faster than collaborative filtering.