Digital Libraries

New Media

Week 13 Min-Yen Kan

New Media

Why important?

Storing knowledge in these media

Communicating about tasks / knowledge

Able to identify how information travels from place to place

New Media to examine:

Instant Messaging

Email

Web logs

Syndication

Wikis

Instant messaging

Synchronous

Like talk and IRC, but centered around user

Buddy list, idle counters, emoticons

Task-based patterns of use:

Mainstream users

Intense users (frequent, more than x conversations)

Continuously logged users (lurking)

Properties of IM

Media switching happens frequently

Used to coordinate F2F meetings, telephone

Easily recordable

Variable presence

Can be anyplace: need location and time for coordination tasks

Idleness hard to determine

Even with manually set “away” features

Lightweight, small footprint

Multitasking frequently

Short conversations

Improving IM

Task-related improvements

means that only some contacts will be active for some tasks

coordination with calendaring

Turn-taking hard to thread when reviewing

More so in multiparty IM

Refactoring may be necessary

More disruptive than email

But can be used as sticky note

Need accurate “ping”

Email – Task-centric

Correlated in business roles

Not just messaging anymore

Has a marked interrupt effect

Jackson 2003 study shows people on average read email right away (within 2 minutes) and take ~ 1 minute to recover from interruption.

Co-opted by many functions needed in information management

Production, transmission and filtering of information

Takes the form of tasks:

Coordination (Time): calendar and deadlines

Collaboration (Other people): contacts

Email – Solutions

Correlated in business roles with the Todo list

One’s own messages as important as others’

Show sent-mail with incoming mail

Tasks need support besides messaging

Email becomes the Personal Information Mangager (PIM)

Email attachments and notes need to be first-class citizens

Attachment synchronization (where’s the most updated version?)

Email – Solutions

Extended responses take a while to write

Show context of response in drafts

Deadlines need to shown to help prioritize

A task involves a limited set of contacts

Use a separate contact list for each specific task

Still need better solutions to identify overviews

Both generic and query-based summaries needed

TaskMaster email client

Finding experts using email

One way: look in email collections for frequent keywords

Another way: view to: and from: as citation link and analyze

One method to combine the two

use HITS algorithm

HITS-based expert finding

Campbell et al. did exactly this (03)

1. Retrieve all emails from group on subject using keywords search (e.g., “digital libraries”)

2. Run HITS on this set of emails to find authorities

3. Assess correlation with human judgment and compare vs. standard tf ranking approach

Limitation:

Need access to emails

Email data needs to be classified to filter noise

Web logs - Blogs

History

web log Þ we blog Þ blog

Blogger et al. (1999): free web publishing

Features

Chronological

Relatively short posts

Frequency

Vocal

Blogs – a public face of the self

Public and private mode simultaneously

Implicit audience makes it more personal than typical web publishing

Usually created for self, family and friends

Something to: remember, share with others, promote, comment

Allows tracking of thoughts in a semi-formal way

Hyper linking ability vital

Filter blogs for knowledge aggregation

Two types of blogs:

Filter: aggregator, work related

Journal: online diaries, personal rants

Filter blogs

Earlier blogs, in which UI emphasized linking

Allowed community to form

Organized by chronology: enforces currency

List other blogs of interest in a blogroll

Blog features

Facilitate community building and awareness

Permalinks

Similar to PURLs

Semi-transparent, with chronological info

http://<username>.company/<username>/<4 digit year>/<2 digit month>/<15 character name>.html

Trackback

Like SGML, automatically know which site links to yours

Implemented by TrackBack ping: a message sent back from one webserver to another.

Content Syndication

Chronological ordering may have spurred it

Want the “freshest” news

Clipping service

Two current standards:

Atom / RSS (Really Simple Syndication)

Allows aggregation of blog items on a single reader / page

Question: How is it different from mailing lists? From news groups?

Wikis – open to the world

Wiki wiki = Hawai’ian for “very quick”

First used in Portland Pattern Repository in 1995

Allows anyone to post or modify pages

Adds edit and create new page buttons to a page

Blurs author and reader

- wikipedia.org

Wiki Properties

Extremely easy to add a link

Use CamelCase

If page with title “CamelCase” doesn’t exist, it will be created as a stub

A collaboration tool for webpages

Currently hampered by non-WYSIWYG editing (need to know HTML)

Navigation and linking difficult

Anarchic link policy too loose

Most sites impose guidelines (although most not enforced)

Recency difficult to see

Refactoring (page restructuring) necessary

Wiki uses and other hazards

Structured knowledge base

Customer support

Reference sites

Digital Libraries?

Skirts issue of trust

Shilling possible

Link spam

Digital Libraries

Analyzing new media

Week 13 Min-Yen Kan

Slide 21

Burstiness in Streams

Tracking ideas through blogs

Strong capabilities of tracking / awareness in blogs

Gruhl et al. envision a similar model for blog idea tracking: infection

Threshold model:

node adopts idea with probability threshold t

Iterate at time t

Cascade model:

If neighbor adopts idea, node adopts with probability p

Topic diffusion in blogs

Topic = keyword

Need to track relevant words w.r.t. time

tf ´ cidf (cumulative idf); corpus is a moving window

Find three distributions of topics

Chatter: topics continuously discussed (e.g., alzheimers)

Spike: topic exhibiting a usage spike, then inactivity (e.g., chibi)

Spiky Chatter: Topics (e.g., microsoft)

Overlay of above two types (multiple spikes possible)

Spike removal possible with spike model

Conclusions

New media allow us to rethink and repackage knowledge and its transmission

Themes of collaboration, informality, recency and ubitiquity throughout along with uncertainty

To think about:

The Virtual Reference Desk is organized as an email triage center. Do you think new media can improve this initiative?

How do the new media types handle the different patterns of use exhibited by scholars? Which tasks are well-supported? Which are not?

References

Bellotti et al. (2003) Integrating tools and tasks: Taking email to task: the design and evaluation of a task management centered email tool Proc. CHI 2003

Kleinberg (2003) Bursty and Hierarchical Structure in Streams Data Mining and Knowledge Discovery, 7(4)

Gruhl et al. (2004) Information diffusion through blogspace Proc. WWW 2004.

Jackson et al. (2003) Understanding email interaction increases organizational productivity CACM

Christopher Campbell et al. (2003) Expertise identification using email communications Proc. CIKM 2003.

Water break

Last break of the year. See ya!

Digital Libraries

Revision

Week 13 Min-Yen Kan

Slide 29

Slide 30

Information Retrieval and Multimedia

Traditional Information Retrieval

Lexicon and posting file construction and compression

Euclidean and cosine similarity

Multimedia

Textual Images: CCITT, OCR sensitivities

Image: vector vs. raster graphics

Audio: perceptual coding for human limitations

Markup Languages

SGML to:

HTML and XML

XML variants: TEI, SMIL, SVG

Indexing and Metadata

Dublin Core addresses all aspects of metadata

Administrative, structural, use, IP and descriptive

Indexing as one part of descriptive metadata

Tradeoff in specificity and exhaustiveness in indexing

Controlled vocabulary

Objectives: distinctive terms, help bridge ASK

Classification

Exhaustive, 1 to 1 mapping of possible subjects

Faceted indexing for faceted metadata

Identifiers

Identifiers

Properties: persistent, unique, fast resolution, decentralized

Two systems: PURL, DOI

OpenURL – solve appropriate copy problem

Bibliometrics

Originated in social networks

Find power laws exponential distributions

Decay in citation rates, impact of time

Co-citation and bibliographic coupling

Centrality (undirected) and prestige (directed)

Applying it to the web:

Pagerank: iterative prestige, rank only

HITS: hubs and authorities on a expanded base set

DL Policy

Economics of the DL

Volume of knowledge vs. publishers’ cost

Search engines acting as marketing;
Websites act as publishing house

Social Aspects

Self-archiving

Preservation: Digital Deposit, Internet Archives

Digital Divide

Rich have access, get richer … poor get poorer

Bridge divide through access to resources and education

Information Seeking

Types of Questions in RI

In contrast to the DL and Web

Seeking as berry-picking

Finding and evaluating sources

Using others: collaborative filtering

Ask-A services and user-user recommender systems

Aspects of seeking

Affective, accessibility and quality factors

Information Chain

And its relationship to citations

Evaluating sources

User Interfaces

HCI goals

Feedback, reduce memory load, scaffolding

Different interfaces for different parts of the seeking process

Query specification, Results display, Relevance feedback

Systems and their properties

VQuery, Filter/Flow, QBIC, Flamenco, Tilebars, Infocrystal, Superbook, Tablelens, Startree, Magic Lens

Patterns of Use

DL, articles have distinct uses

Browsing, searching modes

Particular to user’s role

Web users have limited actions, too

Case study: the “back” button

In both cases, optimize UI to account for these specifics

Applications

Both applications can be structured as a machine learning problem

Recommender Systems

Memory vs. Model

Shilling

Authorship attribution

Non-content word patterns

Duplicate detection

R-measure

New Media

IM, Email, Blogs to Wikis: User based

Purpose and salient characteristics

How do they play a role in the future of the article and the scholar?

Semantic Web: Agent based

Allowing agents autonomy

The web as a giant database

RDF: representing knowledge as triples

OWL: language to map different ontologies

Evaluation

IR based metrics

P / R / Sn / Sp and compound metrics

Library metrics

Use centered vs. materials centered

Micro vs. macro evaluation

Final Exam

1 ½ hours, 20% of final grade

Same format as midterm exam

Definitions

Calculation

Critical essay

Slightly longer (in length) than midterm, questions of higher weight

Emphasizes second half of course

First half still fair game

some questions may need to refer to first half material

Digital Libraries

Presentation Guidelines*

Week 13 Min-Yen Kan

Presentation format & timing

10 minutes of presentation (max 10 slides)

2 minutes (1 slide) to introduce the problem

2 minutes to define the problem

2 minutes evaluation

2 minutes conclusions

The rest is up to you.

5 minutes for questions

Only one group member has to be present

You should be prepared to ask questions of other projects

Not graded, but encouraged

Other details

Will be the same grade for all students unless your team tells me otherwise

Practice at least once

Otherwise, you’ll probably run over time

Anticipate questions

Send me your slides (.PDF or .PPT) to post to IVLE after your presentation

Think about publishing your slides, survey paper on the web to help others

Some presentation guidelines

Introduction:

Involve your audience immediately and throughout the presentation

(1) Tell them what you're going to say, (2) say it, & (3) tell them what you said

Questions:

Carefully listen to questions before answering

Acknowledge the validity of an appropriate question

Don't answer a question that you don't know

Visual aids:

Use 1 figure per minute at most, & 1 figure per 2 minutes at best

Make every figure interesting

Simplify your figures, and then make them simpler.

Explain your figures in detail (including defining axes)

Use figures as a memory (numbers & words) crutch

Don't read from text figures (face audience & paraphrase).

Use a CONCLUSION or SUMMARY figure to show you're done

Overall grading metrics

Oral Presentation Skills:

Correct use of English.

Logical presentation.

Conclusions demonstrate critical thinking.

Emphasize important points.

Good eye contact, do not read presentation.

Appropriate non-verbal communication

Slides:

Make sure your slides are readable.

Use short phrases on slides, say full sentences.

Chose a high contrast color scheme and font (generally sans-serif).

Don’t put too much text on a slide.

Make use of graphics but make sure the graphics do not distract.

Grading metrics

Organization

State what his topic is?

Main point presented clearly?

Speech clearly organized into a few sections?

Scientific Presentation

Cite scientific facts, statistics, statements from authorities?

Use scientific terms and define these terms for the class?

Analysis and Synthesis

Synthesize and compare different articles?

Use of Visual Aids

Visual aids add quality to the presentation?

Sources

Give proper credit to people whose ideas he borrowed?

Figures properly attributed?

Questions

Show respect for those who asked questions?

Understood question?

Answered question well?

Overall Quality

Speaker prepared?

Present adequate information?

Interesting?

Understand the material?

That’s all folks!

Thanks very much!

Hope it has been a fun and worthwhile course for you…


	Why important?
		Storing knowledge in these media
		Communicating about tasks / knowledge
		Able to identify how information travels from place to place

	New Media to examine:
		Instant Messaging
		Email
		Web logs
		Syndication
		Wikis


	Synchronous
		Like talk and IRC, but centered around user
		Buddy list, idle counters, emoticons
	Task-based patterns of use:
		Mainstream users
		Intense users (frequent, more than x conversations)
		Continuously logged users (lurking)


Media switching happens frequently
	Used to coordinate F2F meetings, telephone
	Easily recordable

Variable presence
	Can be anyplace: need location and time for coordination tasks
	Idleness hard to determine
		Even with manually set “away” features

Lightweight, small footprint
	Multitasking frequently
	Short conversations


	Task-related improvements
		means that only some contacts will be active for some tasks
		coordination with calendaring
	Turn-taking hard to thread when reviewing
		More so in multiparty IM
		Refactoring may be necessary
	More disruptive than email
		But can be used as sticky note
		Need accurate “ping”


Correlated in business roles
	Not just messaging anymore
	Has a marked interrupt effect
		Jackson 2003 study shows people on average read email right away (within 2 minutes) and take ~ 1 minute to recover from interruption.
	Co-opted by many functions needed in information management
		Production, transmission and filtering of information
	Takes the form of tasks:
		Coordination (Time): calendar and deadlines
		Collaboration (Other people): contacts


	Correlated in business roles with the Todo list
		One’s own messages as important as others’
		Show sent-mail with incoming mail

	Tasks need support besides messaging
		Email becomes the Personal Information Mangager (PIM)
		Email attachments and notes need to be first-class citizens
		Attachment synchronization (where’s the most updated version?)


	Extended responses take a while to write
		Show context of response in drafts
		Deadlines need to shown to help prioritize

	A task involves a limited set of contacts
		Use a separate contact list for each specific task

	Still need better solutions to identify overviews
		Both generic and query-based summaries needed


	One way: look in email collections for frequent keywords
	Another way: view to: and from: as citation link and analyze

	One method to combine the two
		use HITS algorithm


	Campbell et al. did exactly this (03)
		1. Retrieve all emails from group on subject using keywords search (e.g., “digital libraries”)
		2. Run HITS on this set of emails to find authorities
		3. Assess correlation with human judgment and compare vs. standard tf ranking approach

	Limitation:
		Need access to emails
		Email data needs to be classified to filter noise


	History
	web log Þ we blog Þ blog
	Blogger et al. (1999): free web publishing

	Features
		Chronological
		Relatively short posts
		Frequency
		Vocal


	Public and private mode simultaneously
		Implicit audience makes it more personal than typical web publishing
	Usually created for self, family and friends
		Something to: remember, share with others, promote, comment
		Allows tracking of thoughts in a semi-formal way
		Hyper linking ability vital


	Two types of blogs:
		Filter: aggregator, work related
		Journal: online diaries, personal rants

	Filter blogs
		Earlier blogs, in which UI emphasized linking
		Allowed community to form
		Organized by chronology: enforces currency
		List other blogs of interest in a blogroll


	Facilitate community building and awareness
	Permalinks
		Similar to PURLs
		Semi-transparent, with chronological info

		http://<username>.company/<username>/<4 digit year>/<2 digit month>/<15 character name>.html

	Trackback
		Like SGML, automatically know which site links to yours
		Implemented by TrackBack ping: a message sent back from one webserver to another.


	Chronological ordering may have spurred it
		Want the “freshest” news
		Clipping service
	Two current standards:
		Atom / RSS (Really Simple Syndication)
	Allows aggregation of blog items on a single reader / page

	Question: How is it different from mailing lists? From news groups?


	Wiki wiki = Hawai’ian for “very quick”
	First used in Portland Pattern Repository in 1995

	Allows anyone to post or modify pages
		Adds edit and create new page buttons to a page
		Blurs author and reader
	- wikipedia.org


Extremely easy to add a link
	Use CamelCase
	If page with title “CamelCase” doesn’t exist, it will be created as a stub
A collaboration tool for webpages
	Currently hampered by non-WYSIWYG editing (need to know HTML)
Navigation and linking difficult
	Anarchic link policy too loose
		Most sites impose guidelines (although most not enforced)
	Recency difficult to see
	Refactoring (page restructuring) necessary


	Structured knowledge base
		Customer support
		Reference sites

		Digital Libraries?

	Skirts issue of trust
		Shilling possible
		Link spam


Strong capabilities of tracking / awareness in blogs
Gruhl et al. envision a similar model for blog idea tracking: infection
	Threshold model:
		node adopts idea with probability threshold t
		Iterate at time t
	Cascade model:
		If neighbor adopts idea, node adopts with probability p


Topic = keyword
Need to track relevant words w.r.t. time
	tf ´ cidf (cumulative idf); corpus is a moving window

Find three distributions of topics
	Chatter: topics continuously discussed (e.g., alzheimers)
	Spike: topic exhibiting a usage spike, then inactivity (e.g., chibi)
	Spiky Chatter: Topics (e.g., microsoft)
		Overlay of above two types (multiple spikes possible)
		Spike removal possible with spike model


	New media allow us to rethink and repackage knowledge and its transmission
	Themes of collaboration, informality, recency and ubitiquity throughout along with uncertainty

	To think about:
	The Virtual Reference Desk is organized as an email triage center. Do you think new media can improve this initiative?
	How do the new media types handle the different patterns of use exhibited by scholars? Which tasks are well-supported? Which are not?


	Bellotti et al. (2003) Integrating tools and tasks: Taking email to task: the design and evaluation of a task management centered email tool Proc. CHI 2003
	Kleinberg (2003) Bursty and Hierarchical Structure in Streams Data Mining and Knowledge Discovery, 7(4)
	Gruhl et al. (2004) Information diffusion through blogspace Proc. WWW 2004.
	Jackson et al. (2003) Understanding email interaction increases organizational productivity CACM
	Christopher Campbell et al. (2003) Expertise identification using email communications Proc. CIKM 2003.


Traditional Information Retrieval
	Lexicon and posting file construction and compression
	Euclidean and cosine similarity

Multimedia
	Textual Images: CCITT, OCR sensitivities
	Image: vector vs. raster graphics
	Audio: perceptual coding for human limitations

	Markup Languages
		SGML to:
		HTML and XML
		XML variants: TEI, SMIL, SVG


	Dublin Core addresses all aspects of metadata
		Administrative, structural, use, IP and descriptive
	Indexing as one part of descriptive metadata

	Tradeoff in specificity and exhaustiveness in indexing
	Controlled vocabulary
		Objectives: distinctive terms, help bridge ASK
	Classification
		Exhaustive, 1 to 1 mapping of possible subjects
		Faceted indexing for faceted metadata


	Identifiers
		Properties: persistent, unique, fast resolution, decentralized
		Two systems: PURL, DOI
	OpenURL – solve appropriate copy problem


	Originated in social networks
		Find power laws exponential distributions
		Decay in citation rates, impact of time
		Co-citation and bibliographic coupling
		Centrality (undirected) and prestige (directed)

	Applying it to the web:
		Pagerank: iterative prestige, rank only
		HITS: hubs and authorities on a expanded base set


	Economics of the DL
		Volume of knowledge vs. publishers’ cost
		Search engines acting as marketing; Websites act as publishing house

	Social Aspects
		Self-archiving
		Preservation: Digital Deposit, Internet Archives

	Digital Divide
		Rich have access, get richer … poor get poorer
		Bridge divide through access to resources and education


Types of Questions in RI
	In contrast to the DL and Web
Seeking as berry-picking
	Finding and evaluating sources
	Using others: collaborative filtering
		Ask-A services and user-user recommender systems
Aspects of seeking
	Affective, accessibility and quality factors
Information Chain
	And its relationship to citations
	Evaluating sources


	HCI goals
		Feedback, reduce memory load, scaffolding

	Different interfaces for different parts of the seeking process
		Query specification, Results display, Relevance feedback

	Systems and their properties
		VQuery, Filter/Flow, QBIC, Flamenco, Tilebars, Infocrystal, Superbook, Tablelens, Startree, Magic Lens


	DL, articles have distinct uses
		Browsing, searching modes
		Particular to user’s role
	Web users have limited actions, too
		Case study: the “back” button
	In both cases, optimize UI to account for these specifics


	Both applications can be structured as a machine learning problem
	Recommender Systems
		Memory vs. Model
		Shilling
	Authorship attribution
		Non-content word patterns

	Duplicate detection
		R-measure