TOWARDS MEETING INFORMATION SYSTEMS

Meeting Knowledge Management

Vincenzo Pallotta°, Hatem Ghorbel°, Afzal Ballim

°Faculty of Information and Communication Sciences, Swiss Federal Institute of Technology, Lausanne, Switzerland

Business Processes Services, Japan Tobacco International Inc.

Agnes Lisowska

, Stéphane Marchand-Maillet

School of Translation and Interpretation,

School of Computer Science, University of Geneva, Switzerland

Keywords: Meeting Knowledge Management, Tacit Knowled

ge, Modelling Concepts and Information Integration

Tools, Requirements Analysis, Ontology Engineering, HCI on Enterprise Information Systems.

Abstract: Interaction through meetings is among the richest human communication activities. Recently

, the problem

of building information repositories out of recordings of real meetings has gained interest. We report here a

summary of the first two years of research carried out within the Swiss funded research project (IM)2,

together with some lessons learned and future perspectives.

1 INTRODUCTION

Recently, the problem of building information

repositories out of recordings of real meetings has

gained interest, and several research projects have

started

. The National Centre of Competence in

Research (NCCR) on Interactive Multimodal

Information Management, in brief (IM)2

. (IM)2 is a

network of Swiss research institutions with different

competencies and research traditions, and it is

composed of several individual projects (IPs). The

work reported here, is the contribution to the (IM)2

project by researchers belonging to the Multimedia

Dialogue Management (MDM) and to the

Information Indexing Retrieval (IIR) IPs. A

partnership between (IM)2 project and another

existing project on the same topic, the ICSI Meeting

Broswer project (Morgan et al., 2001) has been

settled. Within this project, a corpus of recorded,

Among the most representative we quote here:

http://www.is.cs.cmu.edu/meeting_room/

http://www.icsi.berkeley.edu/Speech/mr/,

http://www.m4project.org/, http://www.amiproject.org/.

http://www.im2.ch/.

transcribed and annotated meetings has been

collected (Janin et al., 2003) and made available to

(IM)2 partners.

1.1 Meeting Recording Scenarios

The application scenarios envisaged in (IM)2 for

meeting recording, understanding, storage and

retrieval are the following.

• For collabor

ative work, suppose someone missed

one group meeting (new/sick/distant employee)

and needs information about “what happened”

at the meeting.

• For high-level manage

ment, a manager might be

interested in searching the meeting repository as

a whole, by tracking and documenting the

progress of a project over a year, by tracking

and documenting the performance of a

team/employee, or by monitoring the

communication and leadership inside a team.

Recording meetings implies the storage and

ructuring of a large set of heterogeneous

information scattered over time and media. In the

one hand, the raw data, obtained from the various

recording devices, is neither directly usable for the

creation of indexes, nor for content-based access to

the relevant parts of the meeting recordings. On the

464

Pallotta V., Ghorbel H., Ballim A., Lisowska A. and Marchand-Maillet S. (2004).

TOWARDS MEETING INFORMATION SYSTEMS - Meeting Knowledge Management.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 464-469

DOI: 10.5220/0002652204640469

 SciTePress

other hand, meeting minutes are considered as a

fundamental source of information for building

knowledge bases and repositories as pointed out in

(Corrall, 1998), but they are not always available,

especially for informal meetings. For our goals, we

need effective methods to map meeting events to

tractable and accessible information sources. The

type of knowledge contained in meetings is often

referred to as tacit knowledge (as opposed to explicit

knowledge). Tacit knowledge, in order to be useful

to anyone beyond the person who owns it, should be

made available to other people through the process

of communication and sharing, but also converted

into explicit knowledge if that knowledge has to be

reusable for knowledge management. These two

processes are referred to respectively as

socialization and externalization, and can be

fruitfully supported by collaborative technologies

(Romaldi, 2002). The goal of this paper is to

consider collaborative technologies as the

foundation for the design of Meeting Information

Systems.

2 WHAT IS A MEETING?

Following the methodology adopted in (Marchand-

Maillet, 2003), we start by identifying meeting

activities. That is, all possible activities happening

during a face-to-face meeting. Figure 1 presents

these activities as an activity diagram gathered

around a central transition state since we assume that

any combination of activities can be envisaged.

Clearly, the figure shows a factored view of meeting

activities.

Figure 1. Meeting activities

In this paper we will focus more on the abstract

characterization of the knowledge management

process in discussion activities. However, other

multi-modal meeting activities have been discussed

in detail in (Marchand-Maillet, 2003).

2.1 Use cases for Meeting Information

Systems

Users for Meeting Information Systems have been

identified and three possible classes of these users

have been defined:

Participant: a person that is physically

present in the meeting;

Customer: a participant of a project (aware of

the topic), absent from a meeting, or a person

unfamiliar with the project;

Analyst: responsible for the post-processing

of the meeting (e.g. minutes, summaries,

meeting records).

These users may have other roles in the

enterprise as well (e.g. giving a role to a participant

like “manager”).

User's requirements have been initially gathered,

primarily by guessing, and then by classifying

possible user's queries. Browsing is another

modality of access to meeting information systems

not discussed in this paper. The interested reader

may refer to (Marchand-Maillet, 2003). A first

attempt has been proposed in one of the project’s

first deliverable reports (Lalanne and Sire, 2002),

from which some interesting query examples are

classified and reported below.

Situation: Where was taking place the meeting?

Participants: Who were the participants? Who was the

president/moderator of the meeting?

Turn taking: I want to see the turn-taking flow of the overall

meeting. Who talked most during the meeting?

Actions and Events: Which document was projected when X

was talking about topic A? Were there (and when) votes made?

Were there (and when) decisions taken? Were there (and when)

any presentations? Was there any break?

Agenda: What was the agenda of the meeting?

Topics: What were the different topics treated during the

meeting? What did participant A said about topic 1? What were

the questions opened about topic 1? What was the conclusion for

topic 1? What solutions have been chosen concerning topic 1?

Who have accepted solutions X1 concerning topic 1? Which

decisions has been taken concerning topic 1?

Dialogue acts: What were the questions asked concerning topic A

and their corresponding answers?

Tasks: On what issues group members disagreed on? Which

members disagreed and on what subject? What were the tradeoffs

being made and what were the criteria (dimensions) used to make

up a decision? What were the decisions being made? Which

criteria were chosen to take the decision? What were the

competitive issues? Which members were competing against each

TOWARDS MEETING INFORMATION SYSTEMS: MEETING KNOWLEDGE MANAGEMENT

465

other and on which corresponding issues? What information has

been disseminated and by who?

2.2 Elicitation of real user queries

A better characterization of user requirements has

been performed by means of a user queries

elicitation experiment. This study was run in

July/August 2003 to find out what types of things

users want to know about (e.g. people, topics,

decisions, agenda) and how they ask about them

(e.g. language, modalities). Details are available in

(Lisowska, 2003). The experiment was conducted in

the following manner.

A questionnaire was sent and received by e-mail,

which briefly presented the (IM)2 project, explained

the use cases and asked participants to pick one (or

more), and think of questions they might want to ask

to get answers in the specified circumstances. The

following use cases (or scenarios) were chosen in

order to constrain the task, which allows for easier

analysis and acquisition of a more coherent set of

results across participants:

1. A manager tracking employee performance;

2. A manager tracking project progress;

3. An employee reviewing a meeting they

missed;

4. A new employee trying to learn about a

project.

A total of 28 responses were gathered, 14 by

(IM)2 project members and 14 by people external to

the project, resulting in 297 queries. We can

summarize the results of the analysis of the user

queries by highlighting 3 types of knowledge

required in order to interpret the queries and provide

a relevant answer to them:

User models: The fact that people tend to make

assumptions about their dialogue partners based on

various factors and alter their own behaviour

accordingly (e.g. vocabulary, background, goals,

preferences) suggests that a set of representative user

models for the meeting domain is needed.

Domain ontologies: The user, depending on

his/her degree of acquaintance with the domain, is

expected to formulate their queries using a domain-

specific terminology.

Natural language understanding: There are

queries for which various degrees of natural

language understanding are required. In general, the

level considered is semantics and pragmatics, but it

also seems that robust syntactic models are required

in order to extract the right answer type from

queries. Some examples of these types of issues are:

• The use of anaphoric expressions.

• The contextual interpretation of general

concepts.

• The use of indirect questions.

• The use of highly domain-specific abstract

concepts.

The latter seems to be the most difficult issue

since its solution requires the construction of a

meeting ontology and its use in the interpretation of

user queries. For instance, in the query "Who made

constructive criticisms about the proposal?" one

might ask what constitutes a constructive criticism.

To answer this last question the system needs to

know this term has an operational interpretation (e.g.

based on an argumentative model).

2.3 Meeting Data Model

The meeting data model we propose, described in

detail in (Marchand-Maillet, 2003) and in (Pallotta,

2003), assumes that a meeting is part of a project,

held in a given meeting room that will capture the

multimedia data from the meeting room (e.g. video,

audio, shared electronic documents). The data

recorded during the meeting is then stored in parallel

with documents associated with the meeting (e.g.

distributed/shown during the meeting). A meeting

may be structured in terms of its temporal activities

(episodes). These are defined by the meeting states

presented earlier. Minutes, as a set of notes (taken or

not during the meeting) and the text transcripts will

also be stored in this meeting repository. Text

transcripts can be enriched by annotations following

the shallow dialogue model proposed in (Armstrong

et al., 2003).

Similar use case analyses have been carried out

in the context of Electronic Meeting Systems

(EMS). See (Antunes and Costa, 2002) and (Antunes

and Carriço, 2003) for a thorough survey. It is worth

to draw attention here to the fact that our model

must be less constrained and abstract than the EMS

models, since we cannot foresee all possible types of

interactions and processes in face-to-face meetings.

However, the EMS models proposed so far can be

adapted and reused to capture the rationale of

meetings and superimpose a logical structure to

multimedia documents obtained from the meeting

recordings. We will certainly look at those models at

later stages of the project.

2.4 The construction of the meeting

ontology

Ontologies are essential in order to define precise

guidelines for the transcription and annotation of a

large number of recorded meetings, including their

semantic/conceptual annotation (i.e. metadata) and

ICEIS 2004 - INFORMATION SYSTEMS ANALYSIS AND SPECIFICATION

466

the annotation of the dialogue structure. We consider

two types of meeting ontologies:

1. A meeting-type-specific ontology can be used to

represent the information related to the purpose

and nature of the meeting, regardless of the

technical domain in which the meeting takes

place.

2. Domain-specific ontologies complement the

task-related information with structured

knowledge about the domain in which the

dialogue is taking place.

These ontologies must be somehow formalized

and built in a flexible and efficient manner. For both

types of ontologies, their (semi)automatic

production must be also accompanied by

mechanisms that automatically attach the proper

ontologies to the meetings. The attachment will be

obtained by dialogue model detection and the

classification of the different tasks (reporting,

decision making, vote, etc.) that can be observed

during meetings and characterized by specific

interaction patterns. For instance, meeting events

are often introduced by summaries, recaps,

conclusions, sign-offs (i.e. agreements),

disagreements, rejections, etc.

A first rough classification of meeting types has

been considered, where three main classes were

decided upon:

1. Meetings for Executive Strategies. These types of

meetings are aimed at goal formation and are

typically unconstrained. Examples are Board

meetings, Steering committee meetings.

2. Management meetings are aimed at plan

formation for task management (e.g. human

resources and task allocation). They are

generally along the lines of project and staff

meetings and are more structured.

3. Business Processes Meetings are more

structured, disciplined, and often have an

invariant structure (i.e. follow templates). They

are aimed at plan execution (i.e. control).

All the above types of meetings might be part of

another class of meetings characterized by two

essential features: decision making and action point

events. We refer to this class as breakdown

meetings. Action points can be defined as task

assignment with deliverables and delay.

2.5 Argumentative structure of

meetings

In order to answer the types of questions

exemplified in the previous section, we need to

structure the meeting records at a deeper level, by

further annotating them with appropriate meta-

descriptions: "argumentative structures". A simple

but expressive model of an argumentative structure

is the "Issue Based Information Systems" (IBIS)

model, proposed by (Kunz and Rittel, 1970) and

adopted as a foundational theory in some computer-

supported collaborative argumentation (CSCA)

systems such as Zeno (Gordon and Karacapilidis,

1999), HERMES (Karacapilidis and Papadias,

2001), Questmap (Conklin et al., 2001), and

Compendium (Selvin, 2001). We adopt this model

as the reference model for the description of the

argumentative structure of decision-making events

in meetings. The model captures and highlights the

essential lines of a discussion in terms of what issues

have been discussed and what alternatives have been

proposed and accepted by the participants.

2.6 The Meeting Description Schema

The description schema we propose, discussed in

(Ghorbel et al., 2003), is the starting point for the

construction of general meeting model. It is

formalised using XML-schema

and reflects the

substantial aspects of the IBIS model. The Meeting

Description Schema (MDS) is based on the previous

observation that there exist a number of sequencing

regularities in dialogue, adjacency pairs, describing

facts such as, for instance, that questions are

generally followed by answers, issues by solutions,

proposals by acceptances or rejections, etc. In MDS,

the dialogue contexts are represented by

argumentative episodes and can be viewed as snap-

shots of the discussion. When analysing the

dialogue, a single tree structure is not sufficient to

represent the adjacency pairs: consider an answer

that refers to two questions in the discussion. For

this purpose we add a dependency relation

("replies_to"), which links the answer to both the

two questions. The "replies_to" relation induces a

chain structure on the dialogue which is local to

each episode and which enables the visualization of

its context. There is an invariant parametric structure

of discussion episodes which is reported below:

DISCUSS(issue)

PROPOSE(solution/idea/alternative/opinion)

ASK_FOR(explanation/justification)

PROVIDE(explanation/justification)

ACCEPT(explanation/justification)

REJECT(explanation/justification)

ACCEPT(solution/idea/alternative/opinion)

REJECT(solution/idea/alternative/opinion)

http://www.w3.org/XML/Schema/

TOWARDS MEETING INFORMATION SYSTEMS: MEETING KNOWLEDGE MANAGEMENT

467

This structure mirrors, in terms of episodic structure,

the IBIS model.

3 TOWARDS A MEETING QUERY

ENGINE

Searching meeting dialogues poses several problems

when using standard Information Retrieval (IR)

indexing techniques. One important point is that

users may ask different types of queries depending

on their needs, and therefore one single retrieval

strategy may not be sufficient. We believe that

standard text-based IR techniques are partially

adequate to meet the requirements for retrieving

meeting dialogues as we have observed, in the user's

queries elicitation analysis. The link to additional

knowledge (present in the meeting repository in the

form of annotations or links to other knowledge

sources such as, for instance, related documents)

may increase the robustness and performance of the

search engine. We need to improve the efficiency of

a meeting search engine by combining

heterogeneous indexes having different natures

(lexical, semantic) and different modalities (speech,

documents) in the following way:

Indexing techniques:

• thematic information obtained by adapted

indexing techniques;

• semantic/structural information described in

the form of metadata in annotations (e.g.

topic segmentation, dialogue acts,

argumentative structure);

• aligned information (e.g. speech transcription

with related documents);

• knowledge about the structure of the meeting

(e.g. information from the meeting database).

Query interpretation techniques based on deep

linguistic (semantic/pragmatic) analysis of the user's

query in order to identify query expansion

(reformulation) strategies and search strategies (e.g.

selecting the index granularity, selecting the filtering

strategy).

We also propose to consider knowledge-based

methods for flexible query expansion and/or

reformulation. Let's consider for instance a simple

query like:

Who disagreed on issue T?

Following a classical IR approach, one can

imagine enhancing indexing by attaching an

additional "disagreement" term to all the turns

included in episodes which are part of the

argumentative chains induced by the "replies_to"

relation of type:

DISCUSS(issue)

PROPOSE(alternative)

REJECT(alternative).

This solution allows us to have indexed both the

content of the argument (i.e. terms of the episode

DISCUSS(issue)) and the names of the people who

have disagreed (supposing that this information is

indexed for each episodes). However, by collapsing

this information in term-based indexes we loose the

argumentation structure and thus might obtain a

false positive: the person who started the discussion

of issue T is considered to be one of the people who

disagree. Moreover, if there are several

disagreements, the speaker corresponding to a

PROPOSE(alternative) episode can be erroneously

paired with the speaker of the REJECT(alternative)

which is not in the "replies_to" relation. A different

approach, which does not suffer from the above

problems is to answer the first query by gathering all

the episodes with content T of type DISCUSS(issue)

and, for each of the retrieved episodes, following the

argumentative chain to select the associated

PROPOSE-REJECT(alternative) pair.

4 CONCLUSIONS

Multimodal meeting recordings are an extremely

rich source of information, which needs to be

converted into explicit knowledge for its

exploitation in the context of enterprise knowledge

management. Speech technology can help in

supporting extraction of knowledge (McCowan et

al., 2003), and dialogue act disambiguation (Bhagat

et al., 2003). As pointed out in (Brown et al., 2001),

a more formal approach to meeting capture and

analysis, based on a formal theory of argumentation

(e.g. IBIS), must be taken so that the decision-

making process can be more easily traced and

understood.

In this paper we have discussed a possible

foundation of Meeting Information Systems and

proposed a preliminary model based on user

requirements analysis. This work has been carried

out in the framework of the (IM)2 project, which,

reaching its second year, has showed a clear

roadmap for future research.

REFERENCES

Antunes P., Costa C.J., 2002. A Descriptive Framework

for Electronic Meeting Systems Based on the UML

ICEIS 2004 - INFORMATION SYSTEMS ANALYSIS AND SPECIFICATION

468

Language. Technical Report DI/FCUL TR-02-05,

Department of Informatics, University of Lisbon,

Portugal. http://www.di.fc.ul.pt/tech-

reports/abstract.php?report_ref=2002-05.

Antunes P., Carriço L., 2003. Modeling the Information

Structures of Meetingware. In proceedings of

Workshop de Sistemas de Informação Multimédia e

Cooperativos, COOP-MEDIA '03. Porto, Portugal,

October 8, 2003.

http://www.di.fc.ul.pt/~paa/papers/coopmedia-03-

mware.pdf

Armstrong S., et al., 2003. Natural Language Queries on

Natural Language Data: a Database of Meeting

Dialogues. In Proceedings of NLDB'2003,

Burg/Cottbus, Germany.

Bhagat S., Carvey H., Shriberg E., 2003. Automatically

Generated Prosodic Cues to Lexically Ambiguous

Dialog Acts in Multiparty Meetings. In proceedings of

ICPhS 2003 conference, Barcelona, August 2003.

Brown E.W., Cooper J.W., Srinivasan S., Amir A., Coden

A., Poncelon D., 2001. Towards speech as a

knowledge resources. In IBM Systems Journal, vol.

40(4), 2001.

Conklin, J., Selvin, A., Buckingham Shum, S., and

Sierhuis, M. 2001. Facilitated Hypertext for Collective

Sensemaking: 15 years on from gIBIS: Knowledge

Media Insitute, Open University.

Corrall S., 1998. Knowledge Management. Are we in the

knowledge management business? In ARIADNE: the

Web version, issue 18, December 1998,

http://www.ariadne.ac.uk/issue18/knowledge-mgt/.

Gordon, T., and Karacapilidis, N. 1999. The Zeno

Argumentation Framework. In German Artificial

Intelligence 3:20-29.

Janin A., et al., 2003. The ICSI Meeting corpus. In

ICASSP-2003, Hong Kong, April 2003.

Karacapilidis, N., and Papadias, D. 2001. Computer

Supported Argumentation and Collaborative Decision

Making: The HERMES system. In Information

Systems 26, no 4:259-277.

Kunz, W., and Rittel, H. W. J. 1970. Issues as elements of

information systems. Working Paper n. 131,

Universität Stuttgart, Institüt für Grundlagen der

Planung. May 1979.

Marchand-Maillet S., 2003. Meeting Record Modelling

for Enhanced Browsing. Tech. Rep. 03.01, Computer

Vision and Multimedia Laboratory, Computing

Centre, University of Geneva, Switzerland, March

2003.

McCowan I., Gatica-Perez D., Bengio S., Lathoud G.,

2003. Automatic Analysis of Multimodal Group

Actions in Meetings. In IDIAP research report n.

IDIAP-RR 03-27. Martigny, May 2003.

Morgan N., Baron D., Edwards J., Ellis D., Gelbart D.,

Janin A., Pfau T., Shriberg E., and Stolcke A., 2001.

The Meeting project at ICSI. In Human Language

Technologies Conference, San Diego, March 2001.

Lalanne D., and Sire S., 2002. Smart Meeting Minutes

Application Specification. IM2 Technical Report, ver.

2.4, University of Fribourg, Switzerland, August 2002.

Lisowska A., 2003. Multimodal Interface Design for the

Multimodal Meeting Domain: Preliminary Indications

from a Query Analysis Study. Technical Report

IM2.MDM-11, November 2003.

http://www.issco.unige.ch/projects/im2/mdm/docs/M

DMReport-11-AL.pdf

Ghorbel H., Pallotta V., 2003. Argumentative

Segmentation and Annotation Guidelines. Technical

report IM2.MDM-08, June 2003.

http://lithwww.epfl.ch/im2/docs/argument-annot.pdf.

Pallotta V., 2003. A computational dialectics approach to

meeting dialogues tracking and understanding. In

Giacalone Ramat A., Rigotti E., Rocci A. (eds.)

Linguistic and new professions, Special Issue of

Materiali Linguistici vol. 1095.41, Franco Angeli.

Romaldi, V., 2002. Collaborative Technologies for

Knowledge Management: Making the Tacit Explicit?

In proceedings of IS2002 conference on Informing

Science and IT Education. Cork, Ireland June 19-21,

2002.

Selvin, A. et al. 2001. Compendium: Making Meetings

into Knowledge Events. In Proceedings of Knowledge

Technologies 2001, March 4-7, Austin TX.

TOWARDS MEETING INFORMATION SYSTEMS: MEETING KNOWLEDGE MANAGEMENT

469