Michael Dressauer

1. Listening Down the Value
Chain
Using Text-based Predictive
Models to Find New
Opportunities for B-to-B
Businesses
Michael P Dessauer, PhD
PREDICTIVE ANALYTICS WORLD
SAN FRANCISCO 2017

2. 2
Who We Are -
Dow’s Advanced Analytics Group
Translating Data into Profit

3. 3
Where is Dow in the value chain?
Chemical Industry Value Chain
Polymers Specialty Brand
Owner
Consumer
Basic
Chemicals
Upstream Downstream
Traditional communication chain
Dow’s own market listening capabilities
creates a downstream feedback-loop
Market Listening

4. Isolate Dow formulation’s value through consumer sentiment
4
How can Dow generate value from Market Listening?
I really don’t
use my _____.
It doesn’t
really _______
I like my _____
more than _____.
It doesn’t make
me too ______
I love my
_____ because
it’s so _______
when use it!
Brand 1 Brand 2 Brand 3
Product Attribute
Analyze 1000’s of
relevant documents
Search by
• Brand
• Product
• Attribute

5.  Find new market opportunities through Consumer Needs
5
How can Dow generate value from Market Listening?
I really don’t
use my _____.
It doesn’t
really _______
I like my _____
more than _____.
It doesn’t make
me too ______
I love my
_____ because
it’s so _______
when use it!
Analyze 1,000,000’s of
relevant documents
Search by
Products in
market
segment
General and
market-specific
sentiment scoring
Filter on
negative
posts
Identify topic
clusters
Innovations

6. 6
Market Listening Project Process at Dow
Targeted Text
Collection
Entity, Topic, &
Sentiment
Categorization
Modeling &
NLP
Insight
Delivery
Internal & Vendor tools
to search and collect
text
Work with marketing experts
to generate valuable
categorizations
Modeling tools for
categorization and/or
extraction
Deliver interactive
dashboards with
regular data updates

7. 7
Market Insights Services – Finding the right fit.
Best Fit Capability Assessment (< 2 Day)
Dow Business Strategies and Plans
Discovery Workshop (2 Hours)
Self-Service Consulted Assisted Custom
Tell me something I don’t know about a key market
Help me get insight from my value chain Help me understand the dimensions of a target market
1A Commercial Support and Advanced Analytics Collaboration
Self Service
Use appropriate Dow-
licensed tool
Dow Internal Service
Use Dow internal
consulting team’s
expertise
External Consulting
Use when Dow
does not have
internal expertise
External Service
Custom-built
applications combining
internal and external
resources

8. 8
Case Study – Strengthen our Value Proposition
 Brand-specific social media data
sources
 Leveraged several tools for
Sentence parsing
Part-of-speech tagging
Topic identification
 Aspect-specific sentiment
weighting
 Simple delivery

9. Targeted Text Collection
 Domain Expert targeted sources
on Twitter and Facebook
 Data extraction through public
APIs, licensed APIs, and custom
web scraping
 Download to harmonized central
repository
9
Case Study – Relevant Text Collection
Data
Repository
Internal & Vendor tools to
search and collect text

10. Entity, Topic, &
Sentiment
Categorization
 Business Marketing and R&D-lead
effort to determine value case
 Define entities, attributes, and
sentiment to quantitatively assign value
to consumer needs
 Feedback loop with modeling step to
validate results
10
Case Study – Value-driven Text Categorization
Work with marketing experts to
generate valuable
categorizations
Brands?
Product
Attributes?
How to define
sentiment?

11. Modeling & Text NLP
 Natural Language Toolkit (NLTK)
leveraged for sentence level parsing
 SAS Text Miner used to develop
sentiment classes
 Sentences weighted on sentiment
content
11
Case Study – Modeling & Text NLP
Seed
Terms
Positive
Class
Neutral
Class
Negative
Class
Training
Corpus
Modeling tools for
categorization and/or
extraction

12. Insights Delivery
12
Case Study – Deliver Insight
Deliver interactive
dashboards with
regular data updates
Brand 1 Brand 2 Brand 3
Aspect #1
post sentence Topic1Score Topic1Sentiment Topic2Score Topci 2Sentiment
post 1 sentence 1 0.78 -2.33 0.05 1.34
post 1 sentence 2 0.03 1.22 0.88 -1.33
post 1 sentence 3 0.34 0.2 0.33 0
 Simple, Ad Hoc, delivery method
 Each sentence in each post had an
individual sentiment score for each of
the aspects
 Sentiment was rolled up to the Brand
level and reported based on relevance
to the post

13. Thank You
Using Text-based Predictive Models to Find New Opportunities for B-to-B
Businesses
LISTENING DOWN THE VALUE CHAIN
13