Special Topics in Computer Science The Art of Information Retrieval Chapter 10: User Interfaces and Visualization Alexander Gelbukh .

Special Topics in Computer ScienceSpecial Topics in Computer Science

The Art of Information RetrievalThe Art of Information Retrieval

Chapter 10: User Interfaces Chapter 10: User Interfaces and Visualization and Visualization

Alexander Gelbukh

www.Gelbukh.com

2

Previous chapter: ConclusionsPrevious chapter: Conclusions

Inverted files seem to be the best option Other structures are good for specific cases

o Genetic databases

Sequential searching is an integral part of manyindexing-based search techniqueso Many methods to improve sequential searching

Compression can be integrated with search

3

Previous chapter: Previous chapter: Research topicsResearch topics

Perhaps, new details in integration of compression and search

“Linguistic” indexing: allowing linguistic variationso Search in plural or only singular

o Search with or without synonyms

4

Topic “Interfaces”: OverviewTopic “Interfaces”: Overview

Human-computer interaction Search process Support for its parts:

o What to start from

o Query specification

o Showing the results

o Relevance feedback

Support for the whole process

Many specific examples of systemso Long. I’ll omit details

5

Human-computer interaction...Human-computer interaction...

Most important part of the problemo Bad interface kills all the fancy technical features of a

system. If you cannot use it, you don’t care how good it is Chopsticks

A good interface “disappears” -- like air Design principles

o Informative feedback (e.g., relationships) Internal locus of control (= this feedback is customizable)

o Reduce working memory load (e.g., keep track) Permits to return to a temporally abandoned strategy

o Distinct interfaces for novices / experts (simplicity vs power) Example: Google / Advanced search

6

...Human-computer interaction......Human-computer interaction...

Bad interfaces -- Why? I think:

Different weight of tasks foruser and programmer

Programmer: a list of (equal)tasks (functions)

User: Goals. Some tasks arefrequent and some not

Example: Windows XP:o Copy, Print, Delete, publish in Web ...

7

...Human-computer interaction......Human-computer interaction... Information visualizationInformation visualization

Humans are better at images than words Abstract info: more difficult. Interactive mode helps Types

o Brushing and linking: different views of the same info;changing one changes others

o Panning and zooming: Example: clustering.

o Focus and context: fisheye camera

o Magic lenses: temporally change a part of info under lens

Combination: overview plus detailso Hierarchically

8

...Human-computer interaction...Human-computer interaction EvaluationEvaluation

What is evaluated:o the quality of final result (mostly precision, not recall)

o time to learn the system

o time to achieve goals

o error rates

o retention of the use of interface over time

People are very different: what’s good for someis not for others

Difficult to measure and evaluate

9

Information access processInformation access process

Goals. Tool. Tasks Basic interaction model: query result (repeat) Advanced models take into account:

o integration with browsing near-miss is acceptable: use hyperlinks

o Selection of source collection

o Learning while searching. Goal shift while searching

o Getting the info by pieces, not as a set of relevant docs Temporal lateral goals, then return to the main goal

Interface needs to support this

10

Not only searchNot only search

Search is a part of an activity on ... Other tasks of this activity include

o finding trends, making comparisons

o aggregating information, assessing, interpreting, ...

Search is intermitted with them, not a separate subtask Need for a common interface that supports the whole

process. One tool.

11

Topic “Interfaces”: OverviewTopic “Interfaces”: Overview

Human-computer interaction Search process Support for its parts:

o What to start from

o Query specification

o Showing the results

o Relevance feedback

Support for the whole process

12

Starting pointsStarting points

Choosing information source (collection) How to choose? Lists. Hard to guess Overviews: Panning and zooming

o Category hierarchies. Example: Yahoo! Large hierarchies need for good interfaces to navigate them

o Clustering Similarity measures: text, co-citations, co-links (Google), ... Clustering on the fly. Summaries of the clusters. Zooming in. Clustering of search results (combination of search &

navigation) Graphical views like stars or landscapes. Not clear if useful

13

14

15

...Starting points...Starting points

Exampleso Start from some object and correct its desired properties

o Problem: how to find from which one to start?

Dialogueso Model a human librarian. Too complicated

Wizardso Only useful for simple tasks, not IR

Guided tours Automated source selection

o Search in descriptions of collections. Or: meta-search

16

Query specificationQuery specification

Typeso command language?? (problem!),

o form fillin, menu selection, direct manipulation,

o natural language

Problem: people have difficulties using Booleanexpressions (e.g., confuse AND and OR)

A lot of efforts to help the users to (visually)construct what is internally a Boolean query

(cf. Bengt’s talk)

o Query preview helps

17

Presentation of results: context Presentation of results: context

Documents by keywords Result list: document surrogates (detailed/not)

o KWIC – key words in context (kind of abstract for query)

o Now used for Web (e.g., Google)

Full texto Highlight hits in full text

o Tilebars: representation of keyword distribution at a glance

Keywords by documents Helps understanding which keywords are important

18

... Presentation of results: context... Presentation of results: context

Organizing the query resultso Table of Contents

automatically generated trough hyperlinks

o Hyperlink structure

o Tables (but: only two attributes) too little improvement: TableLens

19

20

21

22

Using relevance judgmentsUsing relevance judgments

Relevance feedback: control in terms of the task, notof the machinery that performs it (keyword weights)

Degree of control (over the keywords to include)o Control only (users set the keywords) – worst,

o opaque (reaction only), transparent (users see new words)

o penetrable: users have control over new words – best.

23

Interface for the whole programInterface for the whole program

Example of problem: search window; old resultsversus newly typed query (not executed)

Windows layout: monolithic (simple/little), tiled,overlapping (large/crowded).

Workspaces. Persistent. Use the possibilities of windows

o bad example: Lexis-Nexis

Provide history information. User preferences

24

25

26

Research topicsResearch topics

Many ideas throughout the chaptero some may be obsolete

New interface types! 3D interfaces Ways of assessing the quality of interfaces

27

ConclusionsConclusions

Interface is a key element of the system. If the userscannot use it, it does not matter how good it is.

Interface design choices are important at any stageof the processo Especially to formulate queries

o Also to present results

o 3D interfaces to present results

Also, overall system interface and action tracking Difficult to assess quality. Difficult to find new ideas Very promising if you find them!

28

Thank you!Till December 4 compensation

lecture: December 11, combined with

normal (last) lecture