Digital Libraries

Usability of OPACs and retrieval engines*

Week 9 KAN Min-Yen

* Mostly based on Hearst’s Chapter in
Modern Information Retrieval

What is HCI?

For digital libraries, it is the means of expediting the information seeking process for a human user

“When an interactive system is well-designed, the interface almost disappears, enabling users to concentrate on their work, exploration and pleasure” --- Shneiderman 97

Tenets of HCI

Offer informative feedback

Reduce working memory load

Provide alternative interfaces for novice and expert users

Methods:

Color and Highlighting

Pan / Zoom

Focus + Context

Overview + Details

Evaluating HCI

Ergonomics (human factors)

time to learn

speed of performance

rate of errors

retention over time

subjective satisfaction

ISP Overview

Outline

Searching

Query formulation

Displaying results

Browsing

Categories and Hierarchies

Integrating Frameworks

Query specification

Old: originated from command line interfaces

Suit the system, not the user

e.g. “FIND TW Mt St. Helens AND DATE 1981”

Then: translated for users on OPACs

Subject:

Date: 1980 – 1985 1986-1990

Now: graphical means for query specification

VQuery (Jones 98)

Filter / Flow model

Users can select from the set of attributes and get an appropriate filter widget

(type-in for interest areas, sliders for cost, and buttons for scholarships)

The widget is placed on the screen with flow lines showing ANDs (sequential flow) and ORs (parallel flows).

Water flow dynamically indicates relevant # of items

Against a controlled vocabulary

Lists

Seen on last slide from Filter / Flow

Only good if limited number of entries

Partial Fill-in

Show possible completions of query terms if under a certain number (~ 5)

Re-writing of form fill-in queries

Who is the leader of Sudan? à
Who is the head of state of X (Sudan)?

Tiles (Anick et al. 90)

Use table to represent conjunction and disjunction

Conjunction (AND): rows

Disjunction: columns

Use activation to see query preview from index

Query by Example

Faceted Metadata

Faceted objects give rise to easy methods for summarizing the data

The Flamenco Project http://flamenco.sims.berkeley.edu/cgi-bin/flamenco/arts/Flamenco?username=default

Facet Map
http://facetmap.com

Slide 14

Geographic queries

Geospatial data makes the 2 & 3-D visualization a good metaphor

Geospatial queries

Point-in-polygon

Region

Distance and Buffer Zone

Path

Multimedia

Results display and integration

Keyword in Context (KWIC)

Tilebars (Hearst 95)

Each row represents a topic (conjunction)

The darkness of each tile in a row represent the frequency of occurrence of an item

Infocrystal (Spoerri 95)

Explosion of the Venn Diagram

Uses shape and color to
model to organize
results

More on Infocrystal

Superbook

Searching a software technical manual

Use a expandable table of contents to show results

Shows hits within each level as query preview

Closely modeled by browser in Windows Explorer

12 An Interactive Environment

10 Preface

1 How to Beat the Lottery

1 Tutorial Introduction to

7 Graphical Methods in S

2 Loading in Data

…

3 Building Plots

2 Specialized Plots

2 Advanced Use of S

2 Appendix I

Cha-cha

Brings this paradigm to website searching

Tablelens (InXight)

http://www.tablelens.com

Focus + context to give fisheye distortion to table rows and values

Cartographic Representations

Scatter / Gather

Visualizing Hierarchies

Fisheye zoom
(Furnas 84)

One implementation: Inxight’s StarTree

http://www.inxight.com/
VizServerDemos/demo/
nasa/index.html

Summarization atop IR

Like Superbook, but trades focus for extractive summary (Kan et al. 02)

Relevant: information for summary

Irrelevant: too broad or not on topic,
for broader queries

Intricate: too detailed,
for follow-ups

Merging Document Topic Trees

Norm for a particular type of document

(e.g. travel leaflets, Univ. department descriptions)

Create by aligning topics in trees by similarity

Slide 31

Additional notes on faceted interface

Multiple interfaces possible

Columnar style

Drop down list style

Indented ancestors

Multiviews via Magic Lenses

More than two views of the same space

Lenses show and hide information that isn’t pertinent

Has been applied in query focusing as well

Demo: http://www.memorialhall.mass.edu/activities/media.jsp?itemid=15414&img=0

Bibliometric citations

With citation
rates

Zoom in on the artery

Relevance feedback

Koenenmann and Belkin (96) checked whether relevance feedback helped users

Control: no RF

Opaque: saw new documents

Transparent: can see modifications

Penetrable: can see modifications and change them

Best one

Possible conclusion: RF can help users create better queries as a form of online training

Query History

Relevance feedback as one form

Pros:

Saved searches can help tune a generic system

Decide whether to use advance or naïve interface for person

Start off where one left off

Cons:

Need to track the user

What happens if multiple purposes or used for guests?

(e.g. Buying gifts for a friend)

Conclusions

New systems show advanced text and graphics / animated displays but many are not yet well-evaluated

Tighter integration in query and display

Simultaneous browsing and searching support at all levels

Screen real estate proportions in frameworks dictate useful alternatives

To think about

What types of information do these various UI need to calculate at run-time and what types can be pre-computed?

How do these UI support the tenets we mentioned at the beginning?

Do you feel that HTML / WWW has enhanced UI design or deterred its creativity?

For images and audio:

When is query by example a useful technique and when is it inferior to a search using metadata?


	Usability of OPACs and retrieval engines*
	Week 9 KAN Min-Yen
	* Mostly based on Hearst’s Chapter in Modern Information Retrieval


	For digital libraries, it is the means of expediting the information seeking process for a human user

	“When an interactive system is well-designed, the interface almost disappears, enabling users to concentrate on their work, exploration and pleasure” --- Shneiderman 97


	Offer informative feedback
	Reduce working memory load
	Provide alternative interfaces for novice and expert users

	Methods:
		Color and Highlighting
		Pan / Zoom
		Focus + Context
		Overview + Details


	Ergonomics (human factors)
		time to learn
		speed of performance
		rate of errors
		retention over time
		subjective satisfaction


	Searching
		Query formulation
		Displaying results

	Browsing
		Categories and Hierarchies

	Integrating Frameworks


	Old: originated from command line interfaces
		Suit the system, not the user
		e.g. “FIND TW Mt St. Helens AND DATE 1981”

	Then: translated for users on OPACs
		Subject:
		Date: 1980 – 1985 1986-1990

	Now: graphical means for query specification


	Users can select from the set of attributes and get an appropriate filter widget
		(type-in for interest areas, sliders for cost, and buttons for scholarships)

	The widget is placed on the screen with flow lines showing ANDs (sequential flow) and ORs (parallel flows).

	Water flow dynamically indicates relevant # of items


	Lists
		Seen on last slide from Filter / Flow
		Only good if limited number of entries
	Partial Fill-in
		Show possible completions of query terms if under a certain number (~ 5)
	Re-writing of form fill-in queries
		Who is the leader of Sudan? à Who is the head of state of X (Sudan)?


	Use table to represent conjunction and disjunction
		Conjunction (AND): rows
		Disjunction: columns
	Use activation to see query preview from index


	Faceted objects give rise to easy methods for summarizing the data

	The Flamenco Project http://flamenco.sims.berkeley.edu/cgi-bin/flamenco/arts/Flamenco?username=default
	Facet Map http://facetmap.com


	Geospatial data makes the 2 & 3-D visualization a good metaphor


	Point-in-polygon




	Region


	Distance and Buffer Zone
	Path






	Multimedia


	Each row represents a topic (conjunction)

	The darkness of each tile in a row represent the frequency of occurrence of an item


	Explosion of the Venn Diagram




	Uses shape and color to model to organize results


Searching a software technical manual
Use a expandable table of contents to show results
Shows hits within each level as query preview
Closely modeled by browser in Windows Explorer
12 An Interactive Environment
	10 Preface
		1 How to Beat the Lottery
		1 Tutorial Introduction to
		7 Graphical Methods in S
			2 Loading in Data
			…
			3 Building Plots
			2 Specialized Plots
		2 Advanced Use of S
	2 Appendix I


	http://www.tablelens.com
	Focus + context to give fisheye distortion to table rows and values


	Fisheye zoom (Furnas 84)
	One implementation: Inxight’s StarTree
		http://www.inxight.com/ VizServerDemos/demo/ nasa/index.html


	Like Superbook, but trades focus for extractive summary (Kan et al. 02)
		Relevant: information for summary
		Irrelevant: too broad or not on topic, for broader queries
		Intricate: too detailed, for follow-ups


	Norm for a particular type of document
		(e.g. travel leaflets, Univ. department descriptions)
		Create by aligning topics in trees by similarity


	Multiple interfaces possible

		Columnar style
		Drop down list style
		Indented ancestors


	More than two views of the same space
	Lenses show and hide information that isn’t pertinent
	Has been applied in query focusing as well
	Demo: http://www.memorialhall.mass.edu/activities/media.jsp?itemid=15414&img=0


	Koenenmann and Belkin (96) checked whether relevance feedback helped users

	Control: no RF
	Opaque: saw new documents
	Transparent: can see modifications
	Penetrable: can see modifications and change them
		Best one

	Possible conclusion: RF can help users create better queries as a form of online training


Relevance feedback as one form

Pros:
	Saved searches can help tune a generic system
	Decide whether to use advance or naïve interface for person
	Start off where one left off

Cons:
	Need to track the user
	What happens if multiple purposes or used for guests?
		(e.g. Buying gifts for a friend)


	New systems show advanced text and graphics / animated displays but many are not yet well-evaluated
	Tighter integration in query and display
	Simultaneous browsing and searching support at all levels
	Screen real estate proportions in frameworks dictate useful alternatives


	What types of information do these various UI need to calculate at run-time and what types can be pre-computed?

	How do these UI support the tenets we mentioned at the beginning?

	Do you feel that HTML / WWW has enhanced UI design or deterred its creativity?

	For images and audio:
		When is query by example a useful technique and when is it inferior to a search using metadata?