Taking over routine tasks: Intelligent interfaces for e-mail

60 %
40 %
Information about Taking over routine tasks: Intelligent interfaces for e-mail

Published on October 17, 2007

Author: lrizoli

Source: slideshare.net

Description

A summary of two intelligent interfaces for e-mail, used as a starting point for class discussion.

Presented on Oct. 16, 2007 for CPSC 532B (http://www.cs.ubc.ca/~conati/532b-2007/532-description.html)

Taking over routine tasks Intelligent user interfaces for e-mail Lucas Rizoli 2007-10-16 CPSC 532C

Segal & Kephart, 1999: MailCat: An Intelligent Assistant for Organizing E-Mail Segal & Kephart, 2000: Incremental Learning in SwiftFile Horvitz, 1999: Principles of Mixed-Initiative User Interfaces

Segal & Kephart, 1999:

MailCat: An Intelligent Assistant for Organizing E-Mail

Segal & Kephart, 2000:

Incremental Learning in SwiftFile

Horvitz, 1999:

Principles of Mixed-Initiative User Interfaces

SwiftFile (aka. MailCat ) (Segal & Kephart, 1999 and 2000)

Facilitates message filing Given a message, predicts folder Adds shortcuts to 3 likeliest folders

Facilitates message filing

Given a message, predicts folder

Adds shortcuts to 3 likeliest folders

Segal & Kephart’s goals: “ Substantial benefit to users” “ Users... not required to learn” “ Errors... have no negative impact ” “ User [is] able to ignore it” “ Incremental learning to adapt [to user] ”

Segal & Kephart’s goals:

“ Substantial benefit to users”

“ Users... not required to learn”

“ Errors... have no negative impact ”

“ User [is] able to ignore it”

“ Incremental learning to adapt [to user] ”

Folders predicted using TF-IDF Messages vectors of word frequencies Folders sum of vectors in folder Distance variation of cos distance, SIM4 Closer to folder, likelier to be filed there

Folders predicted using TF-IDF

Messages vectors of word frequencies

Folders sum of vectors in folder

Distance variation of cos distance, SIM4

Closer to folder, likelier to be filed there

1 98 1370 2 72 2415 5 19 7411 user # folders messages # buttons accuracy TF-IDF Prediction Most frequent folders

1

98

1370

2

72

2415

5

19

7411

user #

folders

messages

# buttons

accuracy

TF-IDF Prediction

Most frequent folders

Accuracy with growth and change in mail Calculated using a moving average # messages accuracy

Accuracy with growth and change in mail

Calculated using a moving average

# messages

accuracy

User Model Upward Inference Input User Folders, Filing TF-IDF Calculations Frequency vectors Downward Inference Output MoveTo Buttons Distance calculations

LookOut (Horvitz, 1999)

Nearly automates scheduling Finds scheduling messages using SVM Adds appointment in calendar Notes conflicts If unsure, opens calendar to week Manual, automatic, social-agent Dialogue, speech recognition

Nearly automates scheduling

Finds scheduling messages using SVM

Adds appointment in calendar

Notes conflicts

If unsure, opens calendar to week

Manual, automatic, social-agent

Dialogue, speech recognition

Horvitz’s factors: Value-added Minimizing costs of errors Aware of user’s attention User can start/stop system Socially appropriate behaviour ... “ Substantial benefit ” “ Errors... have no negative impact ” “ User [is] able to ignore ”

Horvitz’s factors:

Value-added

Minimizing costs of errors

Aware of user’s attention

User can start/stop system

Socially appropriate behaviour

...

Uncertainty about goals Best solution given constraints Match actions to certainty User and agent refine results Dialogue to resolve uncertainties History of actions Continued learning “ Incremental learning ”

Uncertainty about goals

Best solution given constraints

Match actions to certainty

User and agent refine results

Dialogue to resolve uncertainties

History of actions

Continued learning

Evidence-based decision model Nothing Dialogue Act

Evidence-based decision model

Nothing

Dialogue

Act

Time-based model of user attention message size (bytes) time before action (sec)

Time-based model of user attention

message size (bytes)

time before action (sec)

User Model Upward Inference Input User Messages, Settings, Responses Messages thru SVM Attention model, Utilities, SVM Downward Inference Output Scheduled appt., Cue for dialogue Decision function

Discussion Where did the principles come from? LookOut How are utilities found? User-set thresholds useful? Which modality is or should be default? Adapting model of attention to situation

Where did the principles come from?

LookOut

How are utilities found?

User-set thresholds useful?

Which modality is or should be default?

Adapting model of attention to situation

SwiftFile Benefits “without demanding anything?” Predictor uses meta-information? Evaluation data was fair sample? Criteria for success were reasonable, right? TF-IDF vs. User-defined rules

SwiftFile

Benefits “without demanding anything?”

Predictor uses meta-information?

Evaluation data was fair sample?

Criteria for success were reasonable, right?

TF-IDF vs. User-defined rules

Add a comment

Related presentations

Related pages

AI, Robotics, and the Future of Jobs | Pew Research Center

Experts envision automation and intelligent ... Robotics, and the Future of ... There will be a labor market in the service sector for non-routine tasks ...
Read more

Outlook Web Access (OWA) for Exchange | Knowledge Center ...

Microsoft Exchange allows users to access and manage their email online through the Outlook Web Access (OWA) interface. OWA provides a full functional ...
Read more

How Technology Is Destroying Jobs | MIT Technology Review

... taking over the tasks that the humans often don ... (Change in Employment and Wages by Skill, Routine ... Have MIT Technology Review delivered ...
Read more

When Robots Take Over Most Jobs, What Will Be the Purpose ...

When Robots Take Over Most Jobs, ... the phenomenon of routine task work disappearing and only the highest and ... Where Robots Are Taking Over.
Read more

The Control Hub of Your Entire Company | Scoro

Scoro is a control hub that lets you access, ... Intelligent Robot. Automates routine tasks. ... This seems to be an invalid email address.
Read more

Project Management Overview - ePMbook

... as well as a challenging routine of ... taking into account the needs ... through existing user interfaces such as web browsers and eMail.
Read more

Home automation - Wikipedia, the free encyclopedia

Home automation is the residential ... a remote interface to home appliances ... to showcase a variety of intelligent home automation ...
Read more

MIS 11 flashcards | Quizlet - Simple free learning tools ...

... Intelligent agents can ... and correcting over and over again ... To automate routine tasks to help firms search for and filter information ...
Read more

Task - definition of task by The Free Dictionary

... usu. routine task, ... And he's much better I verily believe since I said I'd turn over to him the task of finishing the ... Mail; Share: Facebook;
Read more