An Efficient Manifold Ranking Approach for Monolithic Graphs and Semantic Networks

In recent decades, ranking has played a central role in large-scale information retrieval, preference analysis for recommendation systems, and influence analysis on social media. State-of-the-art ranking algorithms, e.g. PageRank, HITS, and their variants, traverse through a large network of connected items and compute their importance (a.k.a. centrality) by considering the incoming and outgoing links of each node. However, since these techniques take into account the centrality as an scalar quantity, it limits our analytical perspective to only one dimension - rank number. In this paper, we introduce an efficient multidimensional ranking approach that offers both node ranking and cluster analysis on any graph-based structure. We demonstrate that our approach is not only compatible with PageRank's scalar centrality but also manifests multi-dimensional spatial distribution of nodes. Therefore, it offers the possibility to perform cluster analysis on graph-based structures and keyword extraction via manifold centrality.