Finding dimensions for text based on heterogeneous information network

Abstract

We propose an approach applicable in the problem of multi dimensions text mining that finds out several sets of phrases which were referred to as the text dimension. Based on the dimensions of text found by the proposed approach, a network could be built by similarities between documents. A method is proposed to transform the network from a coarse-grained one to a fine-grained one. By repeatedly mining phrases sets from the networks of different granularities, we could get a refined text dimensions set. We provide experimental results on text mining showing the computational feasibility and effectiveness for finding text dimensions which combines text mining with network mining and can be used for learning interesting knowledge.