An Automated Tool for Visualizing Large-Scale Online Discussions
1. An Automated Tool for
Visualizing Large-Scale
Online Discussions
The 2nd International Workshop on Agent-based Crowd Decision-
making (AgentCrowd2020) in conjunction with AAMAS2020
Nagoya Institute of Technology, Japan
Yuya Kondo, Ahmed Moustafa, and Takayuki Ito
2. Background
u In recent years, large-scale discussions on the Web have
attracted great attention.
u Large-scale online discussions face several problems.
u One of these problems is that participants need to read a large
number of posts.
u Research on the visualization of online discussions has
been conducted.
3. Aim
u Visualization of online discussions is expected to support
understanding the content of these discussions.
u Creating deeper understanding and new opinions.
u In particular, this paper aims to support the understanding
of specific topics that users are interested in.
4. IBIS Structure
u Issue-based information system[1]
u An approach to structuring the discussion
u Elements of IBIS
u Issues : Need to be answers
u Ideas : Possible answers
u Pros/Cons : Support/Objection to given idea
issues
ideas
questionquestion
object-to
support
respond-to
need to be
answered
possible answers
support or
objection to a given idea
pros or cons
(arguments)
[1]Douglas Noble and Horst WJ Rittel. Issue-based information systems for design. 1988.
5. D3.js
u D3.js is a JavaScript library for rendering dynamic content
in a web browser.
u Use of D3.js allows for flexible expression.
6. Propose Method
u Extracting IBIS structures from
discussion platforms.
u A structured argument becomes a tree
structure.
u Node is a post, edge is a reply.
u A child node is a post related to its
parent node.
u Decide which posts to display based on
the extracted discussion structure
7. Propose Method
u The proposed visualization tool
mainly implements the following
five functions.
u Automatically select visualized posts
u Featured post visualization
u Post label display function
u Post expansion function
u Discussion log display function
8. Experiment Settings
u Theme of Discussion : Medical care, AI technology, Let’s
think about the future of medicine
u The total number of posts : 82
u Number of participants : 21
9. Experiment Settings
u There are two main purposes of the question.
u require understanding of a specific topic
u require comprehensive understanding of the entire discussion
10. Experiment Settings
u Subjects were divided into three
groups.
u Group 1 used D-Agree[2]’s bulletin
board-style discussion display screen.
u Group 2 used the discussion
visualized by the existing method[3].
u Group 3 used the discussion
visualized by the proposed method.
[2] T Ito, D Shibata, S Suzuki, N Yamaguchi, T Nishida, K Hiraishi, and K Yoshino.
Agent that facilitates crowd discussion. 7th ACM Collective Intelligence, pages 13- 14, 2019.
[3] A Kamiya, A Kitagawa, S Shiramatsu, D Shibata, K Yoshino and T Ito.
Prototype of visualization mechanism of ibis structure and progress of discussion to support consensus building
in Web discussion.[Translated from Japanese.] The 81st National Convention of IPSJ, page 41-42, 2019.
11. Evaluation Experimental Results
u Among the questions asking a specific topic, question 1, 3,
and 4 could be answered in less time than other methods.
u we were able to get an answer in question 6 at the same
time as other methods.
12. Conclusion
u Aim
u Automatically visualize large-scale text-based online discussions.
u In particular, support the understanding of specific topics that
users are interested in.
u Propose method
u Visualizing based on the IBIS discussion structure.
u Result
u The proposed approach supports the understanding of the content
of the topic of interest.
13. Future work
u Set to investigate the techniques that enable us to
summarize the content of different posts.