Yamashita 2020 - Overview

Summary

Yamashita, Kanno & Furuta (2020) propose an interactive workshop method for eliciting local causal knowledge from non-experts to build large causal networks for disaster scenario creation. The system combines a three-element causal model (cause, precondition, effect) with a GUI-based workshop procedure and NLP extraction to automate construction of a causal knowledge database.

Research Question and Contribution

Problem: Creating effective disaster drill scenarios requires predicting causal chains of damage events — but non-experts struggle to do this manually, and purely automated NLP approaches lack accuracy and local specificity.

Contribution:

  1. A simplified causal model (cause + precondition + effect) that is more practical than FRAM yet more expressive than simple cause-effect
  2. An interactive GUI-based method for workshops that elicits causal knowledge through structured questioning
  3. NLP techniques (Method A + B) for automatically extracting causal elements from free-text input
  4. Word2Vec-based deduplication to handle paraphrase variation across participants

Published: HCII 2020, LNCS 12217, pp. 437–446. DOI: 10.1007/978-3-030-50334-5_30

Paper Structure

SectionContent
§1 IntroductionProblem motivation, two approaches to knowledge elicitation
§2 Causal ModelCause-precondition-effect model; comparison to FRAM
§3 MethodGUI design, NLP extraction (Method A + B), duplication prevention
§4 Preliminary ExperimentEarthquake workshop with 2 participants; 20 events, 15 preconditions elicited
§5 ConclusionPerformance validation; limitations (no 3+ step chain support)

Key Results

  • NLP verification on 100 sentences: Method A identified 46 correct causal relations, Method B identified 63, combined: 87 (out of 100)
  • Preliminary workshop: 20 events and 15 preconditions (countermeasures) successfully elicited and integrated into database
  • Participants’ local knowledge (e.g., gas stove near carpet) successfully captured and complemented each other

Limitations

  • No support for evoking multi-step causal chains (3+ events deep)
  • Designed and tested for Japanese language only
  • Workshop requires human facilitation

See Also