Syntax Sentiment Analysis Symposium Jeff Catlin
Aug 15, 2015
© 2015 Lexalytics Inc. All rights reserved
SyntaxSentiment Analysis Symposium
Jeff Catlin
© 2015 Lexalytics Inc. All rights reserved 2
Meaning
Semantics + Syntax + Context = Meaning
© 2015 Lexalytics Inc. All rights reserved 3
Semantics
• Definition of a word• Many possible definitions• Dependent on syntax and context
© 2015 Lexalytics Inc. All rights reserved 4
Context
• Who is saying this?• What have they said in the past?• What is the space they’re
talking about
SICK !
SICK !
© 2015 Lexalytics Inc. All rights reserved 5
Syntax
• What we’re going to be focusing on• The effect of sentence structure on the
meaning of a word or phrase.
© 2015 Lexalytics Inc. All rights reserved 6
Simple Example
Billy hit the ball over the house.
© 2015 Lexalytics Inc. All rights reserved 7
Solution
• Humans naturally parse syntax– Billy hit the ball over the house.
• So, learn like a human:– Unsupervised learning across large corpora of text to extract common associations
• Deep learning/Neural Nets• Matrix Factorization
• Bob is going to the store for milk.– You’re not going to see “Milk store closed on Good Friday” in the large corpus – so you
know he’s going to go buy milk.
© 2015 Lexalytics Inc. All rights reserved 8
Syntax-Heavy Examples
Document Sentiment• I was expecting a great experience, but the waiter was awful.• The staff helped me with everything I needed help with, but didn't make me feel helpless.
Entity Sentiment• I love Coca Cola but hate Pepsi. • Apple was doing bad until Steve Jobs returned.• Because Apple was doing bad, Steve Jobs returned.• Apple was doing bad because Steve Jobs returned.• I wish GM created a new, great car.• GM created a new, great car.
© 2015 Lexalytics Inc. All rights reserved
Apple was doing bad because Steve Jobs returned.
9
© 2015 Lexalytics Inc. All rights reserved
Because Apple was doing bad, Steve Jobs returned.
10
© 2015 Lexalytics Inc. All rights reserved
Summary• Semantics + Syntax + Context = Meaning• Many sentences have many valid parses, but that are nonsense for a human• So, use unsupervised learning to understand a valid parse
– John went to the store for milk.
© 2015 Lexalytics Inc. All rights reserved