Autonomous Interface Agents

1
Autonomous Interface Agents

• Henry Lieberman
• Media Laboratory, MIT
• Presented by
• Sumit Taank
• Vishal Mishra

2
Outline

• Introduction to Agents
• Interface Agents
• Autonomous Agents
• Letizia An Autonomous Interface Agent
• Conclusion

3
Introduction to Agents

• No universally excepted definition.
• Libermans definition
• Program that can be considered by the user to
  be acting as an assistant or helper, rather than
  as a tool in the manner of a conventional
  direct-manipulation interface.

4
Properties of Agents

• Learning
• Inference
• Adaptability
• Independence
• Creativity
• Delegating task rather than commanding

5
Interface Agents

• Program that affect the objects in a direct
  manipulation interface, but without explicit
  instruction from the user.

6
Interface Agents

• Likened to robots
• Softbots
• Sensors for the input
• Effectors form the output
• Examples
• Intelligent tutoring systems
• Content-sensitive help systems

7
Autonomous Agents

• Program that operates in parallel with the user.
• Act Independently and Concurrently
• Importance Time Savers, True Delegation
• In Contrast to Conversational Interfaces

8
Autonomous Interface Agents

• Interface Agent may not be Concurrent
• Intelligent tutoring systems
• Critiquing systems
• Autonomous agents maybe non-interactive
• Must have Concurrency, Independence and
  Interactive behavior

9
Letizia Concepts

• Autonomous interface agent for web browsing
• Watches user Web Browsing to learn
• Compiles user profile by content analysis
• Uses TF-IDF
• Searches nearby Web space in parallel
• Offers recommendations independently

10
Letizia Interface

• Consists of 3 Netscape Windows (default)
• Channel Surfing interface style
• Tracks user used window, uses other

11
Traditional Search

• Conversational User must declare interest
  explicitly
• User idle during search
• Search agent idle while user is using browsing
  interface
• Sequential structure, no concurrency

12
Letizia Search

• Treats Web browsing as real time activity
• Considers user attention foremost
• Goal not to retrieve best answer but give
  reasonable recommendation
• in context search
• just in time delivery

13
Users Search

• Browsers tend to encourage a depth-first
  exploration.
• User can get stuck in long fruitless path

14
Users Letizias Search

• Letizia searches breadth-first interleaved with
  users depth-first search.

15
User Profile

• TF-IDF is used to compute content of a document.
• Not reliable but fast.
• User profile is accumulated during browsing.
• Profile may be saved across sessions to make it
  persistent.

16
Recommendation

• Presents recommendations continuously
• Displays recommendation that matches user
  interest.

17
Main Values to User

• Avoiding Dead Ends
• Avoiding Garden Path
• Noticing Serendipitous Connections
• Providing better than nothing guesses when no
  other source of preference

18
Conclusions

• Novel idea to improve web browsing.
• User profile bag of keywords
• User can search on unrelated topics in same
  session.
• Searches only nearby pages.
• Search engines or other web resource can be used
  for search.

19

• ?? QUESTIONS ??