Predicting meeting participants’ note-taking from previously uttered dialogue acts
| Published date | 09 May 2016 |
| Date | 09 May 2016 |
| Pages | 170-185 |
| DOI | https://doi.org/10.1108/JSIT-07-2015-0064 |
| Author | Antje Bothin,Paul Clough |
| Subject Matter | Information & knowledge management,Information systems,Information & communications technology |
Predicting meeting participants’
note-taking from previously
uttered dialogue acts
Antje Bothin and Paul Clough
Information School, University of Shefeld, Shefeld, UK
Abstract
Purpose – The purpose of this paper is to describe a new supervised machine learning study on the
prediction of meeting participant’s personal note-taking from spoken dialogue acts uttered shortly
before writing.
Design/methodology/approach – This novel approach of providing cues for nding important
meeting events that would be worth recording in a meeting summary looks at temporal overlaps of
multiple people’s note-taking. This research uses data of 124 meetings taken from the AMI meeting
corpus.
Findings – The results show that several machine learning methods that the authors compared were
able to classify the data signicantly better than a random approach. The best model, decision trees
with feature selection, achieved 70 per cent accuracy for the binary distinction writing for any number
of participants simultaneously or no writing, whereas the performance for a more ne-grained
distinction of the number of participants taking notes showed only about 30 per cent accuracy.
Research limitations/implications – The ndings suggest that meeting participants take
personal notes in accordance with the utterance of previously uttered speech acts, particularly dialogue
acts about disuencies and assessments appear to inuence the note-taking activities. However, further
research is necessary to examine other domains and to determine in what way this behaviour is helpful
as a feature source for automatic meeting summarisation, which is useful for more efciently satisfying
people’s information needs about meeting contents.
Practical implications – The reader of an Information Systems (IS) journal would be interested in
this paper because the work described and the ndings gained could lead to the development of novel
information systems that facilitate the work for businesses and individuals. Innovative meeting capture
and retrieval applications, satisfying automatic summaries of important meeting points and
sophisticated note-taking tools that suggest content automatically could make people’s daily lives more
convenient in the future.
Social implications – There are wider implications in terms of productivity and efciency. Business
value is increased for the organisation, as human knowledge is built more or less automatically. There
are also cognitive and social implications for individuals and possibly an impact on the society as a
whole. It is also important for globalisation, social media and mobile devices.
Originality/value – The topic is new and original, as there has not been much research on it yet.
Similar work was carried out recently (Murray, 2015; Bothin and Clough 2014). This is why it is relevant
to an IS journal and interesting for the reader. In particular, dialogue acts about disuencies and
assessments appear to inuence the note-taking activities. This behaviour is helpful as a feature source
Our thanks go to the creators of the AMI meeting corpus for providing us with large amounts of
annotated meeting data. We are also grateful to the anonymous reviewers for their valuable
comments.
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/1328-7265.htm
JSIT
18,2
170
Received 21 July 2015
Revised 20 January 2016
Accepted 22 January 2016
Journalof Systems and
InformationTechnology
Vol.18 No. 2, 2016
pp.170-185
©Emerald Group Publishing Limited
1328-7265
DOI 10.1108/JSIT-07-2015-0064
for automatic meeting summarisation, which is useful for more efciently satisfying people’s
information needs about meeting contents.
Keywords Knowledge management, Information management, Data analysis
Paper type Research paper
1. Introduction
Meetings are a vital part of many professional organizations and academic institutions
today. They are the events where information exchange and distribution as well as
knowledge generation and sharing occur. People usually attend many meetings in the
workplace (3M Online Survey, 1998). However, over time, they tend to forget what has
happened in the conversations. To satisfy the information needs of meeting
participants, who cannot remember what happened in a particular meeting that took
place a while ago, or for people who are unable to attend such a gathering at all, a concise
meeting summary is necessary. Traditional minutes are sometimes not sufcient
because they do not record every detail; above all, they cannot capture emotions and
certain discussion elements such as what led to a decision (Whittaker et al., 2008;Renals,
2010). In the business world, it is time-consuming to interrupt co-workers on missed or
forgotten meeting content; thus, innovative automatic meeting information capture and
retrieval systems are likely to improve employees’ productivity (Benson and Standing,
2008).
It is now easily possible to record meetings at low cost and store them online or in a
corporate network, but there is usually too much information available to quickly search
for what users require to know. To overcome this problem of information overload,
meeting browsers (Wellner et al., 2004;Castronovo et al., 2008;ICT Results, 2010) have
been developed to display a better overview of a recorded meeting, such as audio, video,
speech transcripts, presentation slides, summaries, keywords or personal notes. This
facilitates the storage and retrieval of important meeting information and improves
corporate memory by providing a better record of such multi-party conversations. The
automatic creation of meeting summaries particularly enhances the performance of
meeting browser environments because this approach saves time, effort and money, as
opposed to producing handcrafted documents about the most informative meeting
events. This also increases the productivity, as manual minutes are expensive to create
and sometimes incomplete.
Recently, there has been a growing interest in automatic meeting summarization
(Buist et al., 2004;Yu and Nakamura, 2010;Wang and Cardie, 2013;Murray, 2015a).
Dialogue act (DA) types play a vital role in the meeting discussion, as they are usually
meaningful, longer connected parts of the language that express speech acts. Weigand
(2016) describes speech acts in more detail. Such utterances usually contain the
speaker’s intentions and communicative goals. Generally, they can be seen as a suitable
information source for summary-worthy utterances in meetings (Wrede and Shriberg,
2003;Hsueh and Moore, 2007). In particular, important decisions and action items occur
when certain DAs are uttered in meeting discussions. Thus, it is encouraging to examine
whether the occurrence of such DA types is linked to other meeting activities, for
example, note-taking.
This paper presents a supervised machine learning study that explores the
relationship between DAs and multiple meeting participants’ note-taking shortly after
171
Meeting
participants’
note-taking
To continue reading
Request your trial