The path towards Semantic Email: Summary and Outlook Simon Scerri, Brian Davis, Siegfried Handschuh
[email protected]
Copyright 2008 Digital Enterprise Research Institute. All rights reserved.
www.deri.ie
Enhanced Messaging Workshop AAAI 2008 July 2008, Chicago
Main Contribution • Goals – Support for the correct Interpretation, Handling and Keeping Track of email messages – Improvement to Email Visualisation, Personal Information Management and Workflow Management
• Results – Semanta: Plug-ins for email clients – Semi-automatic Email Annotation – (Semantic) Knowledge integration on Social Semantic Desktop
• Primary lesson learnt – Although email usage is overly flexible, THERE ARE clear common trends.
2
Tools and Techniques • Method - Exposing implicit knowledge to machines re: – Artefacts of email communication process (people, tasks..) – The Evolution of the communication (thread structure) – Ad-hoc Workflows taking place within email (information requests, meeting scheduling..)
• Techniques used – Semantic Web technologies, including the sMail ontology – Content annotation models based on Speech Act Theory
• Tools & Applications – GATE: General Architecture for Language Engineering – Developed Semanta for Outlook/Thunderbird – Integrated on the Social Semantic Desktop [SSD]
3
?
Challenges • Modelling Email Action Items (Requests, Assignments..) Role Initiative Requestive
Continuative Completive
Informative
Noun
Negotiative
Imperative Propose
Activity Deliver
Request Assign
Action
4
Abort
Data
Decline Event
Suggest
Object
Task
Information
Feedback
Resource
Sender
Both Recipient
Subject
Challenges • Modelling Email Action Items (Requests, Assignments..) Role Initiative Requestive
Continuative Completive
Informative
Noun
Negotiative
Imperative Propose
Activity Deliver
Decline
Request Assign
Action
Data
Event
Abort
Task
Suggest
Object
Information
Feedback
Resource
Sender
Both Recipient
Subject
Example “...Please make sure you have the document ready!..” Assign
5
Task
Recipient
Challenges • Modelling Email Action Items (Requests, Assignments..) • Modelling Ad-hoc email workflows
6
Challenges • Modelling Email Action Items (Requests, Assignments..) • Modelling Ad-hoc email workflows • Semi-Automatic classification of Email Action Items
7
Challenges • • • •
8
Modelling Email Action Items (Requests, Assignments..) Modelling Ad-hoc email workflows Semi-Automatic classification of Email Action Items Implementation (Semanta) & Integration with SSD
Challenges • • • •
9
Modelling Email Action Items (Requests, Assignments..) Modelling Ad-hoc email workflows Semi-Automatic classification of Email Action Items Implementation (Semanta) & Integration with SSD
Evaluation • Evaluation of Email Speech Act Model – – – –
Measuring Inter-annotator Agreement for manual annotation κ-statistic between 0.623 and 0.811 (-1< κ<1) Marginally better than earlier model despite more parameters Experiment & Results published at LREC
• Evaluation of Semi-Automatic Content Annotation – Comparison of “Gold Standard” Manual Annotation with Automatic Annotation in GATE – In Progress...
• Evaluation of Semanta – Features, Added Benefits, Ease of use (User Interface) – To be done…
10
Related Work & Implications • Related Work – Speech Act & Email • Carvalho et. al – Continuation • Lampert et. al – Similar modelling ideas
– Semantic Email • Dowell et. al – Semantic Email Processes
• Implications of Work – Benefits Email Classification, Visualisation, Retrieval, Personal Information Management.. – Semi-automatic annotation of electronic conversations – Information Integration with SSD knowledge
11
Open Questions • Email Ad-hoc Workflows
?
Can Speech Acts within Emails be chained? To which extent?
• Semi-automatic annotation of content
? ?
Cost of wrong compared to correct annotations? Worth it? Statistical vs KB vs ML vs Mixed approaches. Best technique?
• Semanta
? ?
Non-invasive client extension with full-feature support? Evaluation: Qualitative vs Quantitative – what can be measured?
• Collaboration
?
12
Any areas for ‘symbiotic’ collaboration?