Hidden Space and Segmented Labelling in Text Visualization

DOI

10.2991/assehr.k.200207.061How to use a DOI?

Keywords

hidden space, segmented labelling, feature reduction, text visualization, model interpretation

Abstract

Text visualization can interpret large size documents with various linguistic units and glyphs. Each unit owns its advantages of intuition and precision, which could be visualize under different space efficiencies. For example, histogram of word frequency is an intuitive glyph but not precise, and word embedding could be optimized globally but not intuitive. Previous studies have applied many linguistic units and glyphs with implicit combinations, but lack an approach to align those units explicitly to sustain interpretability and predicability. In another side, ever growing methods of feature reduction and selection require a framework to compare and interpret hidden spaces with regard to large volume documents. To align and visualize linguistic units intensively, we proposed a visualization method to interpret document with its distribution on continuous space. Also, the accuracy of the segmented labelling is compared with the min-max and entropy methods. The result shows that: 1) our visualization is flexibility and efficiency to exhibit large volume documents; 2) of feature selection accuracy, the segmented labelling has comparable advantage on various parameters of hidden space.

Copyright

© 2020, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Xiaoguang Zhu
AU  - Xin Cai
AU  - Peiyao Nie
PY  - 2020
DA  - 2020/02/17
TI  - Hidden Space and Segmented Labelling in Text Visualization
BT  - Proceedings of the International Academic Conference on Frontiers in Social Sciences and Management Innovation (IAFSM 2019)
PB  - Atlantis Press
SP  - 389
EP  - 400
SN  - 2352-5398
UR  - https://doi.org/10.2991/assehr.k.200207.061
DO  - 10.2991/assehr.k.200207.061
ID  - Zhu2020
ER  -