Mining Text Enriched Heterogeneous Citation Networks

摘要

The paper presents an approach to mining text enriched heterogeneous information networks, applied to a task of categorizing papers from a large citation network of scientific publications in the field of psychology. The methodology performs network propositionalization by calculating structural context vectors from homogeneous networks, extracted from the original network. The classifier is constructed from a table of structural context vectors, enriched with the bag-of-words vectors calculated from individual paper abstracts. A series of experiments was performed to examine the impact of increasing the number of publications in the network, and adding different types of structural context vectors. The results indicate that increasing the network size and combining both types of information is beneficial for improving the accuracy of paper categorization.