Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Network-based Topic Structure Visualization

Yeseul Jeon, Jina Park, Ick Hoon Jin, Dongjun Chungc

TL;DR

The paper addresses the difficulty of analyzing inter-correlated topics by proposing a Gaussian latent-space item-response model (LSIRM) applied to a topic–words matrix derived from Biterm Topic Model outputs. It simultaneously estimates latent topic positions and uses a Procrustes alignment and oblique rotation to enable intuitive, two-dimensional topic visualization, while a composite word-score $s_{i,j}$ integrates topic probabilities and latent distances for interpretable word selection. Applied to COVID-19 PubMed literature, the method reveals three coherent topic clusters—outbreak/impact, symptoms/treatment, and molecular-body processes—and demonstrates robustness of topic interpretation across different word-set sizes. The approach offers a practical, interpretable framework for exploring topic networks without relying on ad hoc similarity measures, with publicly available software and data for replication.

Abstract

In the real world, many topics are inter-correlated, making it challenging to investigate their structure and relationships. Understanding the interplay between topics and their relevance can provide valuable insights for researchers, guiding their studies and informing the direction of research. In this paper, we utilize the topic-words distribution, obtained from topic models, as item-response data to model the structure of topics using a latent space item response model. By estimating the latent positions of topics based on their distances toward words, we can capture the underlying topic structure and reveal their relationships. Visualizing the latent positions of topics in Euclidean space allows for an intuitive understanding of their proximity and associations. We interpret relationships among topics by characterizing each topic based on representative words selected using a newly proposed scoring scheme. Additionally, we assess the maturity of topics by tracking their latent positions using different word sets, providing insights into the robustness of topics. To demonstrate the effectiveness of our approach, we analyze the topic composition of COVID-19 studies during the early stage of its emergence using biomedical literature in the PubMed database. The software and data used in this paper are publicly available at https://github.com/jeon9677/gViz .

Network-based Topic Structure Visualization

TL;DR

The paper addresses the difficulty of analyzing inter-correlated topics by proposing a Gaussian latent-space item-response model (LSIRM) applied to a topic–words matrix derived from Biterm Topic Model outputs. It simultaneously estimates latent topic positions and uses a Procrustes alignment and oblique rotation to enable intuitive, two-dimensional topic visualization, while a composite word-score integrates topic probabilities and latent distances for interpretable word selection. Applied to COVID-19 PubMed literature, the method reveals three coherent topic clusters—outbreak/impact, symptoms/treatment, and molecular-body processes—and demonstrates robustness of topic interpretation across different word-set sizes. The approach offers a practical, interpretable framework for exploring topic networks without relying on ad hoc similarity measures, with publicly available software and data for replication.

Abstract

In the real world, many topics are inter-correlated, making it challenging to investigate their structure and relationships. Understanding the interplay between topics and their relevance can provide valuable insights for researchers, guiding their studies and informing the direction of research. In this paper, we utilize the topic-words distribution, obtained from topic models, as item-response data to model the structure of topics using a latent space item response model. By estimating the latent positions of topics based on their distances toward words, we can capture the underlying topic structure and reveal their relationships. Visualizing the latent positions of topics in Euclidean space allows for an intuitive understanding of their proximity and associations. We interpret relationships among topics by characterizing each topic based on representative words selected using a newly proposed scoring scheme. Additionally, we assess the maturity of topics by tracking their latent positions using different word sets, providing insights into the robustness of topics. To demonstrate the effectiveness of our approach, we analyze the topic composition of COVID-19 studies during the early stage of its emergence using biomedical literature in the PubMed database. The software and data used in this paper are publicly available at https://github.com/jeon9677/gViz .
Paper Structure (10 sections, 4 equations, 2 figures, 1 table)

This paper contains 10 sections, 4 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Data
  4. Estimating Topic Relationships Using Latent Space Item Response Model
  5. Scoring the Words Relation to Topics
  6. Application
  7. Topic-words matrix of COVID-19 Biomedical literature using BTM
  8. Words selection for each topic based on the score
  9. COVID-19 Topic Structure Visualization
  10. Conclusion

Figures (2)

  • Figure 1: If topics are located in the center in latent position (a), two distinct subgroups are observed in the score and affinity plot (b). Otherwise, such subgroup patterns are not observed in the score and affinity plot (c). In (b) and (c), orange color (id of $1$) indicates top 20% of words of high score words while green color (id of $0$) indicates the remaining words.
  • Figure 2: Topic architecture visualization of COVID-19 literature with three topic clusters (A) - (C). The arrow indicates the direction of latent positions by increasing the words size. (A) COVID-19 and its impacts on different areas. (B) Symptoms and treatment of COVID-19 and relevant studies. (C) Molecular studies associated with the COVID-19 infection.