Post on 10-Feb-2017
Learn to Develop
Custom Text Classifiers
with MeaningCloud
October 5th, 2016
Webinar
Custom Text Classifiers with MeaningCloud
Before we get started…
Presenter
How to participate
• Send questions with the chat feature, or
• Click the “Raise your hand” button to speak
and we’ll enable your mic
• Afterwards, you’ll be able to access a recording of the
webinar and its contents as tutorials on our blog
Antonio Matarranz
CMO
Custom Text Classifiers with MeaningCloud
The purpose of this webinar…
To learn how to implement
the highest-quality
text clasifiers
for your application
Custom Text Classifiers with MeaningCloud
Agenda
Introduction to text classification
Text classification in MeaningCloud: API, built-in
models
Why to develop custom classifiers
Text classification technologies: machine learning,
rule-based, hybrid
Development of classification models in
MeaningCloud: tools, process
Practical case
Conclusions and Q&A
Custom Text Classifiers with MeaningCloud
Text classification
Uses a predefined taxonomy = {categories}
Associate each text to one or several categories
Economy
Technology
Society
Sports
Taxonomy Class-
ification
Model
Custom Text Classifiers with MeaningCloud
Classify: What? What for?
Text classification is the Text Analytics most extended task
Text
News
Web content
Contact center
Surveys
Social posts
Medical records
Application
Tag & Enrich
Targeted advertising
Analyze interaction
Interpret feedback
Understand conversation
Statistics
Custom Text Classifiers with MeaningCloud
MeaningCloud: “Meaning as a Service”
Sign up, and use it for FREE at
http://www.meaningcloud.com
Custom Text Classifiers with MeaningCloud
Text analytics, in the cloud (and on-premises)
Extract meaning and actionable insights from unstructured content
Automation of costly manual activities
MeaningCloud provides this service as a convenient, web-based offering
OpinionsFacts
Concepts
Organizations
People
Semantic
Analysis
Relationships
Themes
Custom Text Classifiers with MeaningCloud
MeaningCloud’s APIs
Identifies occurrences of
names of people,
organizations, abstract
concepts, quantities, etc.
Theme classification
according to
predefined taxonomies
Identifies general and
attribute-level polarity
Distinguishes among 60
languages
Detailed morphosyntactic analysis Evaluates the impact of
opinions on several
reputational axes
Discover meaningful topics and
similarities among texts without
relying on predefined
taxonomies
Custom Text Classifiers with MeaningCloud
MeaningCloud: standard classification models
‘Out-of-the-box’ support of
well-known predefined
classification standards
IPTC: news
IAB: targeted advertising
EuroVoc: public
administration
Social media: social
conversations
… and more to come
https://www.meaningcloud.com/developer/documentation/supported-models
Custom Text Classifiers with MeaningCloud
Add-in for Excel
An experience fully integrated into Excel
Easy to use - No programming!
The most convenient way to evaluate, prototype, and use MeaningCloud
11
Custom Text Classifiers with MeaningCloud
Let’s focus on a practical example
Yelp Reviews of restaurants in London
Custom Text Classifiers with MeaningCloud
WE NEED AN EASY AND
POWERFUL WAY TO DEVELOP
CUSTOM CLASSIFIERS
Standard models are not enough
Custom Text Classifiers with MeaningCloud
14
Example: custom classification for financial
services
The insights
that we need to
extract define
the classification
model
Custom Text Classifiers with MeaningCloud
MeaningCloud customization tools
Custom Text Classifiers with MeaningCloud
Classification technologies
Classifiers use patterns/vectors that represent each category
Technologies to generate those representations
• Statistic
• Rule-based
Training
documents
for category
Machine
learning
Rules for
category
Rule
codifier
Rule 1
Rule 2
Rule 3
Rule 4
Category
representation
Category
representation
Custom Text Classifiers with MeaningCloud
Comparison between classification technologies
Statistical Rules
Benefits Fast, provided tagged texts
are available
Good accuracy for long texts
No false positives
Very good accuracy for non
extensive environments
Disadvantages “Black box” approach
False positives difficult to
correct
Bias in results, depending on
training
Costly if starting from scratch
False negatives, depending on
rule quality
Difficult to scale
Requires deep domain knowledge
Hybrid approach: best of both worlds
Custom Text Classifiers with MeaningCloud
Defining a new category:
hybrid approach
Rule-based
Training-based
Possible to opt for one of
the approaches, or to
combine both, depending
on the application
Custom Text Classifiers with MeaningCloud
Defining a category: training
Fed with precodified training texts
Based on machine learning technology
Training text
Custom Text Classifiers with MeaningCloud
Defining a category: rules
Compulsory terms Forbidden terms
Terms that increase relevance Terms that decrease relevance
Custom Text Classifiers with MeaningCloud
An iterative method to develop custom classifiers
Understand problem
Define taxonomy
Configure categories
Validate classification
Validate insights
Yes
End
Yes No
No
Examples
Rules
+ Examples
+ Rules
Modify
taxonomy
Custom Text Classifiers with MeaningCloud
Our goal: world cuisine taxonomy
Custom Text Classifiers with MeaningCloud
Rule example: American Food category
Custom Text Classifiers with MeaningCloud
Where to get ideas for rules
Domain knowledge,
manual definition
Use n-gram extractor on
manually tagged texts
Extract information from manually
tagged texts using other tools, e.g.,
• Topic extraction
• Text clustering
Lists of terms featured in category-specific documents, e.g., Wikipedia
articles
Custom Text Classifiers with MeaningCloud
Conclusions
The highest-quality text classification, at
your fingertips
Hybrid technology combines development
speed and accuracy
Customization tools allow to develop models
without programming
Agile development process, e.g.,
Configuration – Excel validation cycles
Machine
learning
Rules
Custom Text Classifiers with MeaningCloud
Democratizing the extraction of meaning
High quality semantic analysis
Optimized technology mix
Continuously updates semantic resources
High-level APIs, e.g., User Profiling
Customizable to customer domain: models, dictionaries, sentiment
Affordable, risk-free
Mature, tested technology
Test and use for FREE (40,000 requests per month)
Pay per use
No commitment or permanence
Commercial plans for all needs
For developers and non
technical users
Add-in for Excel
Standard web services APIs
Plug-ins and SDKs for diverse environments and languages
Plug-and-play approach
OpinionesTemasHechos
Conceptos
Organizaciones
Personas
Relaciones
Custom Text Classifiers with MeaningCloud
Q & A
Custom Text Classifiers with MeaningCloud
Stay tuned to our emails and blog
We’ll be posting a recording of the webinar and
its contents (data and models) as tutorials soon!
Custom Text Classifiers with MeaningCloud
Thank you for your attention!
Questions, suggestions...
Antonio Matarranz
CMO
amatarranz@meaningcloud.com
http://www.meaningcloud.com