Responsive Information Architect Responsive Information Architect - - PowerPoint PPT Presentation

responsive information architect responsive information
SMART_READER_LITE
LIVE PREVIEW

Responsive Information Architect Responsive Information Architect - - PowerPoint PPT Presentation

Responsive Information Architect Responsive Information Architect Context- -Sensitive Information Seeking Sensitive Information Seeking Context Zhen Wen Zhen Wen In collaboration with Vikram Vikram Aggarwal Aggarwal, Keith Houck, , Keith


slide-1
SLIDE 1

Responsive Information Architect Responsive Information Architect

Context Context-

  • Sensitive Information Seeking

Sensitive Information Seeking

Zhen Wen Zhen Wen

In collaboration with In collaboration with Vikram Vikram Aggarwal Aggarwal, Keith Houck, , Keith Houck, Shimei Shimei Pan, Pan, James Shaw, and James Shaw, and Michelle Zhou Michelle Zhou

  • Dept. of Intelligent Multimedia Interaction
  • Dept. of Intelligent Multimedia Interaction

IBM T. J. Watson IBM T. J. Watson http:// http://www.research.ibm.com www.research.ibm.com/RIA /RIA

slide-2
SLIDE 2

2

Input Challenges Input Challenges

Systems often have limited capability of Systems often have limited capability of understanding user inputs understanding user inputs

1440 Broadway, NYC 1440 Broadway, NYC Train stations within 10 blocks Train stations within 10 blocks

Crown Plaza at UN

slide-3
SLIDE 3

3

Output Challenges Output Challenges

One One-

  • size

size-

  • fits

fits-

  • all output is context

all output is context-

  • insensitive

insensitive

“ “hotels hotels” ”

Honeymooner Honeymooner

room services room services luxury amenities luxury amenities

Conference planner Conference planner

business amenities business amenities capacities capacities location . . . location . . .

Family vacation planner Family vacation planner

room type room type child child-

  • care facilities

care facilities… …

slide-4
SLIDE 4

4

Output Challenges Output Challenges

1440 Broadway, NYC 1440 Broadway, NYC

Fragmented output is ineffective Fragmented output is ineffective

Restaurants within 10 blocks Restaurants within 10 blocks

slide-5
SLIDE 5

5

What Went Wrong What Went Wrong

Data management / search engines Data management / search engines

Text Docs Text Docs Images/Videos Images/Videos Structured Data Structured Data

Querying Querying Results Results

Request Request Response Response

slide-6
SLIDE 6

6

Data management / search engines Data management / search engines

Text Docs Text Docs Images/Videos Images/Videos Structured Data Structured Data

Query Query Results Results

Request Request Response Response

Conversation User Environment Data Context Context-

  • sensitive

sensitive Input Interpretation Input Interpretation Context Context-

  • tailored

tailored Response Creation Response Creation

Exploit user interaction context to improve user Exploit user interaction context to improve user information information-

  • seeking experience

seeking experience

Our Approach Our Approach

Responsive Information Architect (RIA) Responsive Information Architect (RIA)

slide-7
SLIDE 7

7

RIA Demo RIA Demo

  • Real-estate application

− Exhibit the difficult problem of info-seeking

  • Hospitality application

− Demonstrate RIA reusability

slide-8
SLIDE 8

8

RIA Key Technologies RIA Key Technologies

Request Request Response Response Context Management Context Management

Conversation Conversation User User Environment Environment Data Data Context-driven Input Interpretation Context-tailored Presentation

Data Management Data Management

slide-9
SLIDE 9

9

RIA Key Technologies RIA Key Technologies

Request Request Response Response Context Management Context Management

Conversation Conversation User User Environment Environment Data Data Context-driven Input Interpretation Context-tailored Presentation

Data Management Data Management

slide-10
SLIDE 10

10

Context Context-

  • Sensitive Interpretation

Sensitive Interpretation – – Using Language Cues Using Language Cues

Previous query: colonial homes Previous query: colonial homes just those near IBM Hawthorne just those near IBM Hawthorne

1.

  • 1. term segmentation

term segmentation

slide-11
SLIDE 11

11

An Example: Context An Example: Context-

  • Sensitive

Sensitive Interpretation Interpretation

Previous query: colonial homes Previous query: colonial homes just those near IBM Hawthorne

Ref op = ‘<=’ val = “1” Constraint 1: companyName = ‘IBM’ Constraint 2: companyLoc = ‘Hawthorne’ stationName = ‘Hawthorne’ cityName = ‘Hawthorne’

Constraint: style = ‘colonial’ Constraint: style = ‘colonial’ House House

near

  • 3. use context to attach
  • 3. use context to attach

different pieces different pieces

  • 2. semantic
  • 2. semantic labeling

labeling

Constraint 1 Constraint 1 Company Company Train Station Train Station City City Constraint 2 Constraint 2

Ref Ref

located-in

slide-12
SLIDE 12

12

  • Diverse expressions for the same thing

Diverse expressions for the same thing

1. 1. Pull that house off my favorite list Pull that house off my favorite list 2. 2. Delete this house Delete this house 3. 3. I I’ ’d like to remove the green one here d like to remove the green one here

  • Fuzzy expressions

Fuzzy expressions

1. 1. Show me houses in Show me houses in good good school district school district 2. 2. Find me Find me cheap cheap hotels hotels

  • Concepts/expressions unknown to systems

Concepts/expressions unknown to systems

1. 1. Show houses with Show houses with fences fences 2. 2. I am new here and I work in Hawthorne. I would like to find I am new here and I work in Hawthorne. I would like to find houses close to my company houses close to my company

However However… …

System interpretation capability System interpretation capability is/will always be limited is/will always be limited User expressions User expressions

System System capability capability

slide-13
SLIDE 13

13

1. 1. Teach users to adapt to system capability Teach users to adapt to system capability

– – Map user inputs to valid system Map user inputs to valid system-

  • acceptable inputs

acceptable inputs in context in context

U U S S

Adaptive Interpretation Adaptive Interpretation

q1 q1 q2 q2 q q

slide-14
SLIDE 14

14

1. 1. Teach users to adapt to system capability Teach users to adapt to system capability

– – Map user inputs to valid system Map user inputs to valid system-

  • acceptable inputs

acceptable inputs in context in context

2. 2. Teach system to adapt to new user inputs Teach system to adapt to new user inputs

– – Extend system capability Extend system capability

U U S S

q q q1 q1 q2 q2

Adaptive Interpretation: Our Adaptive Interpretation: Our Approach Approach

slide-15
SLIDE 15

15

An Example: User Adaptation An Example: User Adaptation

q q: Show : Show tudor tudor houses in houses in good good school districts school districts q q’ ’: Show : Show tudor tudor houses in school districts with over houses in school districts with over 95% of students attending college 95% of students attending college

< <q q, , q q’ ’> >

Store Store

slide-16
SLIDE 16

16

Our Key Technologies Our Key Technologies

Request Request Response Response Context Management Context Management

Conversation Conversation User User Environment Environment Data Data Context-driven Input Interpretation Context-tailored Presentation

Data Management Data Management

slide-17
SLIDE 17

17

Output Creation Output Creation

1.

  • 1. Decide output content and form at run time

Decide output content and form at run time 2.

  • 2. Update system output incrementally

Update system output incrementally Automatically create system output tailored to a Automatically create system output tailored to a given user request in context given user request in context

4+ 4+ bedrm bedrm, 2+ bath , 2+ bath colonials in cities in the colonials in cities in the north along Hudson north along Hudson I found 9 houses satisfying your criteria I found 9 houses satisfying your criteria

Technical challenges Technical challenges

slide-18
SLIDE 18

18

Output Creation Output Creation

1.

  • 1. Decide output content and form at run time

Decide output content and form at run time 2.

  • 2. Update system output incrementally

Update system output incrementally Automatically create system output tailored to a Automatically create system output tailored to a given user request in context given user request in context

4+ 4+ bedrm bedrm, 2+ bath , 2+ bath colonials in cities in the colonials in cities in the north along Hudson north along Hudson I found 9 houses satisfying your criteria I found 9 houses satisfying your criteria

Technical challenges Technical challenges

slide-19
SLIDE 19

19

Find houses in Chappaqua Find houses in Chappaqua (16) (16) Desktop w/o graphics Desktop w/o graphics PDA PDA

Examples: Environment Relevant Examples: Environment Relevant

slide-20
SLIDE 20

20

RIA Summary RIA Summary

Enable a context Enable a context-

  • sensitive interaction paradigm to

sensitive interaction paradigm to assist users in info assist users in info-

  • seeking

seeking

Understand user inputs in context Understand user inputs in context Provide tailored information presentation in context Provide tailored information presentation in context

slide-21
SLIDE 21

21

Ongoing & Future Work Ongoing & Future Work

  • Extend to unstructured information

Extend to unstructured information

– – Text docs, images/videos Text docs, images/videos

  • Extend information seeking to

Extend information seeking to information analysis information analysis

– – Leverage data summarization, mining Leverage data summarization, mining techniques in context techniques in context

slide-22
SLIDE 22