Deakin University
Browse

File(s) under permanent embargo

A new evolving tree-based model with local re-learning for document clustering and visualization

Version 2 2024-06-06, 08:08
Version 1 2017-04-05, 11:41
journal contribution
posted on 2024-06-06, 08:08 authored by WL Chang, KM Tay, Chee Peng Lim
The Evolving tree (ETree) is a hierarchical clustering and visualization model that allows the number of clusters to grow and evolve with new data samples in an online learning manner. While many hierarchical clustering models are available in the literature, ETree stands out because of its visualization capability. It is an enhancement of the Self-Organizing Map, a famous and useful clustering and visualization model. ETree organises the trained data samples in the form of a tree structure for better presentation and visualization especially for high-dimensional data samples. Even though ETree has been used in a number of applications, its use in textual document clustering and visualization is limited. In this paper, ETree is modified and deployed as a useful model for undertaking textual documents clustering and visualization problems. We introduce a new local re-learning procedure that allows the tree structure to grow and adapt to new features, i.e., new words from new textual documents. The performance of the proposed ETree model is evaluated with two (one benchmark and one real) document data sets. A number of key aspects of the proposed ETree model, which include its topology representation, learning time, as well as recall and precision rates, are evaluated. The results show that the proposed local re-learning procedure is useful for handling increasing number of features incrementally. In summary, this study contributes towards a modified ETree model and its use in a new domain, i.e., textual document clustering and visualization.

History

Journal

Neural processing letters

Volume

46

Pagination

379-409

Location

Berlin, Germany

ISSN

1370-4621

eISSN

1573-773X

Language

eng

Publication classification

C1 Refereed article in a scholarly journal

Copyright notice

2017, Springer

Issue

2

Publisher

Springer