A Feature-Enhanced Ranking-Based Classifier for Multimodal Data and Heterogeneous Information Networks

We propose a heterogeneous information network mining algorithm: feature-enhanced Rank Class (F-Rank Class). F-Rank Class extends Rank Class to a unified classification framework that can be applied to binary or multiclass classification of unimodal or multimodal data. We experimented on a multimodal document dataset, 2008/9 Wikipedia Selection for Schools. For unimodal classification, F-Rank Class is compared to support vector machines (SVMs). F-Rank Class provides improvements up to 27.3% on the Wikipedia dataset. For multimodal document classification, F-Rank Class shows improvements up to 19.7% in accuracy when compared to SVM-based meta-classifiers. We also study 1) how the structure of the network and 2) how the choice of parameters affect the classification results.