A New Approach to Real Time Intent and Sentiment Analysis
Aloke Guha Kapil Tundwal
Sentiment Analysis Symposium March 5-6, 2014, New York
Cruxly
Sentiment?
Why business cares about intents?
Find Customers Build Sell Support
Why intent is hard to detect?
Grammar
Punctuation Spelling Sarcasm
How
• Grammar-aware parsing • Verb classification • Real-time detection • Horizontal first . . . vertical later
Source-Agnostic
• SMS, Emails, Social Media • Mobile apps, Location-based services
Intent Detection Basis*
Event Detection
Text Extraction
Email / IM
Social Media
Web Posts
Date Location
Names Ext. Opinion
. . .
Event Detection
Logic
Event Signals
Tokenization Segmentation
Text Content
Sentence Phrase
Text Units
Parser
Grammar Rules
Event Definition
Natural Language Processing
Ref: USPTO 20120245925, “METHODS AND DEVICES FOR ANALYZING TEXT,” Guha, Kireyev, Lampert, and Tundwal, 2012
Under the Hood (Twitter case)
Tweets (Keywords/KIP)
Requests (Keywords)
Tweet ID + Intent Signal
(PostgresSQL)
Tweets Content Store (DynamoDB)
Cruxly Intent Detection (AWS)
Reports / Dashboard Tracker Editor
(web app)
Aloke Guha: Analytics Drives Big Data Drives Infrastructure, 29th IEEE MSST 2013
Analytics: Event / Intent Detection
Source/Device Metadata: Poster,
Date/Time, #Followers,
Location, . . .
User Metadata: Keyword / KIP
Custom: RT / Ad Hoc Query
Tweets (Keywords)
Streaming API Client
Examples
Intent Summary
Intent Summary: Comparison Across Brands
Inte
nt: B
uy
Leads
Inte
nt: L
ike
Inte
nt: D
islik
e
Inte
nt: Q
uest
ions
/Req
uest
s
Geo-Location
Intent for Iterative Analysis
Future Work
• Better polarity – orthogonal to grammar rules • ‘Activation’ (accept, agree, etc.) verbs* • Increase depth analysis • Different grammars – other languages
*B. Levin, English verb classes and alternations: a preliminary investigation, 1993, University of Chicago Press
Conclusions
• Actionable intent and event detection • Grammar-aware parsing to add semantic basis • Real-time response with optimized analysis • Vertical applications
Selected References
1. USPTO #20120245925, “Methods and Devices for Analyzing Text,” Guha, Kireyev, Lampert, and Tundwal, September 27, 2012
2. A. Guha, “Analytics Drives Big Data Drives Infrastructure,” Keynote presentation, 29th IEEE Mass Storage Conference, May 2013.
3. B. Levin, English verb classes and alternations: a preliminary investigation, 1993, University of Chicago Press.
4. A. Esuli, S. Baccianelli, and F. Sebastiani, “SentiWordNet 3.0: An enhanced lexical resource for sentiment analysis and opinion mining,” Proc. 7th Conf Intl.’ Language Resources Evaluation, May 2010.
5. E. Cambria, C. Havasi and A. Husain, “SenticNet 2: A semantic and effective resource for opinion mining and sentiment analysis,” Proc. FLAIRS Conf., 2012
6. A. Gangemi, et al, “Frame-Based Detection of Opinion Holders and Topics: A Model and a Tool,” IEEE Computational Intelligence, Feb. 2014