Download presentation
Presentation is loading. Please wait.
Published byRoss Morgan Modified over 9 years ago
1
Carnegie MellonCarnegie Mellon PMA: A Mobile Context-Aware Personal Messaging Assistant Senaka Buthpitiya Deepthi Madamanchi Sumalatha Kommaraju Martin Griss CyLab Mobility Research Center Mobility Research Center Carnegie Mellon Silicon Valley 1
2
Carnegie MellonCarnegie Mellon Agenda Introduction to Email Sorting Related Work PMA – Design and Architecture Experiments & Results Conclusion Future Work 2
3
Carnegie MellonCarnegie Mellon What is a “Mobile Context-Aware Personal Messaging Assistant”? An advanced rule-based email management system which uses the mobile user’s context and email content to classify emails prioritize emails selectively deliver key messages to mobile phone Uses real-time context information from : hard sensors (GPS, accelerometer, etc.) on Mobile phone soft sensors (calendar, …) 3
4
Carnegie MellonCarnegie Mellon Email Flooding in the Real World Busy professionals receive in excess of 50 emails per day, 23% require immediate attention 13% require attention later 64% are unimportant Problem is even worse for mobile professionals Difficult to sort through emails on mobile devices Wastes precious bandwidth and battery life End Result: Wastes time sorting through unwanted emails Drastic reduction in productivity! 4
5
Carnegie MellonCarnegie Mellon Problems Most email sorting/classification programs take only email-content into account Depending on users’ contexts, the emails that they wish to see vary Depending on the users’ contexts the number of emails they can scan through varies Email sorting/classification programs consider importance only Importance and urgency are orthogonal yet affects email sorting equally 5 UnimportantImportant Non- Urgent Evite for a BBQ. From manager: Client visit pushed back by another month. Urgent Online auction: you were out bid. Son missed his bus, pick him up from school.
6
Carnegie MellonCarnegie Mellon Related Work A Personal Email Assistant, Bergman et. al., 2002 CoolAgent: Intelligent Digital Assistants for Mobile Professionals, Griss et. al., 2002 Combining Bayesian and Rule Score Learning Automated Tuning for SpamAssassin, Seewald, 2004 6
7
Carnegie MellonCarnegie Mellon PMA Architecture PMA separately rates emails according importance and urgency using context information and email content e.g. – email from the user’s boss about present meeting is important and very urgent 7 PMA decides on what-to deliver, how-to-deliver and where-to-deliver according to user’s context e.g. – deliver as SMS, text-to-voice SMS, forward to co-worker Uses a rule-based system for decision making
8
Carnegie MellonCarnegie Mellon Context Information Gathered from hard sensors on a Nokia N95 (which also doubles as a delivery point for selected emails) Gathered from soft sensors such as Google Calendar Context includes all information related to user including, Static context such as name and family details Dynamic context such as meeting topic, driving speed User preferences 8
9
Carnegie MellonCarnegie Mellon Experiment - 1 AIM – Test effectiveness of PMA’s urgency and importance classifiers For various user contexts, PMA classifies a test set of emails separately for importance and urgency compared against ratings for the same emails by user 9 Number of type X emails correctly classified by PMA Precision = Number of emails classified by PMA as X Number of type X emails correctly classified by PMA Recall = Total number of emails selected by users as type X
10
Carnegie MellonCarnegie Mellon Summary of precision and recall of importance classification Summary of precision and recall of urgency classification Results 10 RandomPMA Recall33.3%96.3% Precision26.1%88.2% RandomPMA Recall8.3%94.8% Precision8.3%92.6%
11
Carnegie MellonCarnegie Mellon Experiment - 2 AIM – Test effectiveness of PMA’s delivery agent and overall system For various user contexts, PMA decides on what action to perform with a given email SMS to user Send to users as text-to-voice SMS Folder for later viewing Take no action compared against user’s expected action on each email 11
12
Carnegie MellonCarnegie Mellon Results 12
13
Carnegie MellonCarnegie Mellon Conclusions PMA sorts and delivers messages that are relevant to the user in his current context, effectively Uses emails content and user’s context information for decision making PMA uses separate scales to measure urgency and importance of an email PMA is scalable for all inbox sizes PMA is easily personalized to suit the requirements of any user for better accuracy 13
14
Carnegie MellonCarnegie Mellon Future Work Performance of PMA Machine learning schemes to automate the learning from user feedback Improve run-time Generalization of PMA Support for various email accounts Yahoo! mail, Hotmail, etc. Support for additional message types (SMS, IM, RSS, etc.) Personalization of PMA User interface to create/edit custom rules Mobile device interface for feedback and usability 14
15
Carnegie MellonCarnegie Mellon Thank You 15
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.