The document discusses envisioning a discussion dashboard to help understand collective intelligence from web conversations. It aims to orient readers to discussions by highlighting controversial parts and providing a framework to integrate manual analysis tools. The dashboard would summarize discussions using a 5W model - who, what, when, where, why. It would identify knowledge-based versus emotionally charged claims and how purpose-related keywords differ. Understanding perspectives is important for reusing or balancing opinionated discussions.
5. Challenges
Digital Enterprise Research Institute www.deri.ie
n Discussion isn’t limited to a single platform,
discussion space, or argumentation tool.
n How do we identify & summarize disagreement
on the Web?
n Can we highlight the points of contention?
Minority opinions?
5
6. Dashboard goals
Digital Enterprise Research Institute www.deri.ie
Orient readers to discussions.
Spot juicy/tough parts automatically.
Provide a framework for integrating with manual tools
for analysis & sensemaking.
6
7. 5W’s – a simple, familiar model
Digital Enterprise Research Institute www.deri.ie
Who
What
When
Where
Why
7
13. Purpose-related keywords
Digital Enterprise Research Institute www.deri.ie
Knowledge
statistics
Values
truth
secret
Rhetoric
you can thank
Judgment/Opinion
eradicate
tough
rejecting
13
14. Purpose matters
Digital Enterprise Research Institute www.deri.ie
Knowledge-oriented discussions are straightforward
to reuse
Opinion-oriented discussion types may require
caveating or balancing
emotion makes a discussion more interesting
can also indicate the potential for bias.
14
16. Acknowledgments
Digital Enterprise Research Institute www.deri.ie
Thanks to our collaborators!
Katie Atkinson, Trevor Bench-Capon, Adam Wyner (Liverpool)
DERI Social Software Unit
Rhetorical Structure, W3C Health Care and Life Sciences
Funding
Science Foundation Ireland Grant No. SFI/08/CE/I1380
(Líon-2)
Short-term scientific mission (STSM 1868) from the COST
Action ICO801 on Agreement Technologies
16
22. When
Digital Enterprise Research Institute www.deri.ie
Message order
Out of date
Superceded
‘First mover advantage’
‘Last word’
Timelines
Depth of the reply network
Tends to indicate what has been most heavily discussed
22
23. Where
Digital Enterprise Research Institute www.deri.ie
Geographic information
Assumed viewpoints
Likely biases
Genre and source
How big are individual messages?
Do they stand on their own? Require reply context?
How fast does the conversation move?
23
24. Genre & source matters
Digital Enterprise Research Institute www.deri.ie
24
25. Context to combine these?
Digital Enterprise Research Institute www.deri.ie
25
26. Digital Enterprise Research Institute www.deri.ie
What are the turning points in a discussion?
Which viewpoints have diverse support?
What justifications are given for viewpoints?
26
30. Transform Debates into
Argument Frameworks
Digital Enterprise Research Institute www.deri.ie
(1) Households should pay tax for
their garbage.
(4) (1)
Paying tax for garbage increases
recycling, so households should Arrow: premise
pay.
(3) (1)
Wyner, van Engers, & Bahreini.
Recycling more is good, so people From Policy-making Statements
to First-order Logic.
should pay tax for their garbage. EGOV 2010
30
31. Calculate best options
(non-contradictory opinions)
Digital Enterprise Research Institute www.deri.ie
Wyner, van Engers, & Bahreini.
From Policy-making Statements to First-order Logic.
EGOV 2010
31
Notas del editor
http://events.kmi.open.ac.uk/cscw-ci2012/
For instance, can we automatically detect discussions with high levels of disagreement? Can we understand the minority views? Either to persuade them to join the majority, or to find the wisdom in their alternate points of view? Can we identify explanations and justifications? “ The minority is always right.” - Henrik Ibsen
Highlightingg variance to Identify high variance and minority opinions. Bring summaries together. Show variance and minority opinions. Detect explanations for beliefs. Develop a framework for integrating machine and hand-summaries. Look at argumentation and persuasion structures.
Conceptually simple Salient
detect the prevalence of knowledge, emotion, and values as a first approximation to the purpose. High sentiment and low sentiment messages can be found through sen- timent analysis [21], which we also use as a first indication of whether people agree and how strongly their views are expressed. Values are abstract qualities such as utility, beauty, respect, and patriotism; these can be found with gazetteers. Knowledge-based discussions often cite statistics, experts, and studies, which can be text-mined; they may also commonly use argumentation schemes such as expert opinion.
Detecting the purpose of the discussion… Using keywords and rhetorical analysis Provides context
Adding funding/collaborators slide.
attach statements to the people and organizations Avoid bias show the reply networks between pairs of people, perhaps indicating authors with icons or avatars.
Forum and listserv posts often rely on the surrounding messages for coherence and context. Temporal aspects of a discussion are also related to the genre; [11] observed that listservs have short, intense exchanges, organized as a tree, while blogs promote slower diffusion and may have multiple ancestor posts. Thus the source genre is relevant when combining messages from different genres, and in some cases messages may not be understandable without the context of the other messages to which they reply. Geographic information may serve a purpose in serious discussions, for instance, in contentious discussions about placenames; since naming conventions tend to be heavily correlated with geographic ties [27], indications about such ties can provide context for moderators about who is taking part in the discussion.
Web source: Combining messages from different genres must take the source into account. Time, tree, context.
They don’t say how they extracted these – but they say Someone makes statement (1) Someone else gives (4) as a reason/premise for (1) Someone else gives (3) as an additional reason for (1) (2) Is a counterproposal with a range of supporting reasons === Icons: http://findicons.com/icon/27954/girl_5?id=27964# http://findicons.com/icon/27930/boy_8?id=27939# http://findicons.com/icon/27955/girl_4?id=27965#