Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Developing novel methods for reviewing audio records of meetings
University of Dublin, Trinity College, Ireland.ORCID iD: 0000-0003-3211-6529
2010 (English)In: 15th Annual Conference of the Health Informatics Society of Ireland (HISI), ICS HISI , 2010Conference paper, Published paper (Refereed)
Abstract [en]

One of the limiting factors in the development of an audio-visual meeting record for

multidisciplinary team meetings is the problem of easy assess to information in the

recordings afterwards. Multiparty meeting browsing and retrieval is an active

research area, and this research reports our effort in segmenting a full meeting into

units representing independent patient case discussions.

We propose a novel solution to segment patient case discussions (PCDs) from

multidisciplinary medical team meetings (MDTMs) based on "content-free" features.

Logistic regression model shows that the probability of a vocalisation leading a PCD

is related to participant's role and vocalisation duration. In developing the theoretical

models, and testing the concepts, a content-free topic segmentation method was

conducted with AMI meeting corpus. Classification schemes are indentified that serve

as practical approaches for topic segmentation.

A good segmentation algorithm should have similar quantities of segments as

reference, and positions of topic boundary should match reference. So Pk, WD as well

as balance factor are applied as evaluation metric, instead of classification accuracy.

These three metrics bridge the gap between segmentation and classification.

Vocalisation data are naturally imbalanced and sequentially related, which challenge

many classifiers. Proportional threshold naive Bayes classifier and Boosting with

decision tree base classifier overcome such problems and brings results comparable

with text-based segmentation. With these classifiers, simple conversational features

(including vocalisation pauses and overlaps) work as indicators of topic shifts. We

also find that the extent of topic homogeneity influences segmentation results.

Place, publisher, year, edition, pages
ICS HISI , 2010.
National Category
Engineering and Technology
Research subject
Information Systems
Identifiers
URN: urn:nbn:se:kau:diva-56968OAI: oai:DiVA.org:kau-56968DiVA: diva2:1120223
Conference
Health Informatics Society of Ireland
Available from: 2017-07-05 Created: 2017-07-05 Last updated: 2017-07-05

Open Access in DiVA

No full text

Search in DiVA

By author/editor
Kane, Bridget
Engineering and Technology

Search outside of DiVA

GoogleGoogle Scholar

CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf