Dimensionality Reduction and Graph Embedding Of Non-Relational Data

Abstract

In the past decade there is an enormous growth of high dimensional data made the analysis of the data more difficult. Texts are the predominant source of information on the global scale. As there are many text mining techniques available Bag-of-Words is a standard approach. A drawback of this model is that the semantics and structure of the text are not preserved. Graphs have very interesting properties which address these issues in BoW model. Most of the graph analytics suffers from high computational cost and space cost. So, there is a need for transform the graph into low dimensional vector space. Graph embedding is an efficient way to address this issue. In this paper we made a detailed study of types of input graph and available graph embedding techniques.