Applications of NLP to become a high earning ML Engineer.pdf

NLP Applications

1.  
LearnTube|CareerNinja
Applications of NLP
Nanda Kishore Mallapragada

2. What is NLP

3. Applications of NLP
●Language Translation
●Next sentence prediction
●Sentiment Analysis
●Chatbot Q/A
●Grammar correction
●Voice monitoring devices -
Alexa, Google Assistant
●Classifying emails.
●Sentence Classification
●Named Entity Recognition
●Search Engines
●Time series forecasting
●OCR Applications.
●Document reading and
classification.
●Resume Shortlisting

4. Applications of NLP

5. Applications of NLP - Named Entity Recognition

6. Applications of NLP - Image Captioning

7. Prominent Use cases of NLP
●Customized Email Generation
●Routing the Leads
●Customers’ feedback analysis

8. Customized Email Generation
●Today, we understand the overall performance and effectiveness of a campaign/email.
However, attribution of campaign effectiveness & performance is not available at a digital asset
granularity.
●Marketing is unable to serve relevant content to end customers – do not have data on who
should be served what.
●To serve end customers with relevant and personalized content, it is important to understand
the customer’s interest/engagement not just with the holistic campaign, but also at the level of
each piece of content.

9. Customized Email Generation
Feature Extraction is done using
FaceNet (A pre-trained deep
convolutional network) It is a face
recognition system. It is a system that,
given a picture of a face, will extract
high-quality features from the face and
predict a 512-element vector
representation these features, called a
face embedding.
Lifestyle image
Products:
Inspiron 15 and
XPS Desktop
Woman
~28 yrs. oldAsian
Dominant Color
Code: ‘cebbac’
Keywords –
‘new computer
purchase today’,
‘exceptional
experience’, ‘Sleek
designs’,
‘sensational
performance’
Text is best
understood by
university
graduates.
Number of
words: 217
Reading
time: ~75
sec
Visual Features
Textual Features
. . .
Extraction of more textual features are
being explored e.g. capturing all large
bold sized text, which also make up the
headline in html file, subject statements
of campaign emails, CTA text etc.

10. Customized Email Generation
●Raw text Cleaning
●Normalizing Text
●Removing Stopwords
●Tokenization
●Stemming
●Lemmatization
●POS Tagging

11. Customized Email Generation
●Normalizing Text
●Text normalization is the process of transforming text into a single canonical form that it might not have had before.
●Example: INDIA is my Country – india is my country
●Tokenization
●Tokenization is the first step in NLP. It is the processing of splitting a string into small units or tokens which can
be used for further analysis.
●Example: INDIA is my Country – “INDIA”, “is”, “my”, “country”
●Removing Stopwords
●NLP libraries contain predefined stopwords like (the, to, of, was, a, an ….). These stopwords can be removed from a
sentence.

12. Customized Email Generation
●Stemming
●Stemming is the process of converting the tokens into their root word or their word stem.
●Example: the stem of the word eating, eaten, eats, etc. is eat.
●Lemmatization
● Lemmatization does morphological analysis of the word and returns the base word known as a lemma. What this means
is lemmatization makes use of the dictionary to identify the root word rather than just cutting off suffix and prefix.
●Example: going, went, gone – go (root word – lemma)
●POS (Parts of Speech) Tagging
●It is the process of mapping each word in a sentence to a particular part of speech (Noun, Verb, Adverb, Adjective,
etc.)
●Example: I(Pronoun) want (Verb) an (Determiner) early (Adjective) upgrade (Noun).

13. Customized Email Generation
●Keyword Extraction
●WordClouds
●ngrams
●Rake
●Spacy
●Yake

14. Customized Email Generation
WordClouds

15. Customized Email Generation
Unigrams

16. Customized Email Generation
Bigrams

17. Routing the Leads
•The goal is to categorize the comments received from Dell Forms into four classes using Machine Learning.
–Inbound Sales
–Small Business
–Support
–Spam
•The process flow of class identification



•We manually tagged ~5000 comments for Deep Learning (Bert, Roberta and Xlnet approaches) based
classification of “Sales” and “Support”.

Comments
Identify
“Spam”
Comments
and filter out
Rule
Based
Classify remaining
Comments to
“Sales” and
“Support”
DL methods
Rule
Based Classify “Sales”
Comments to “Small
Business” and
“Inbound”

18. Customer Feedback Analysis
●Sentiment Analysis
●Analyze the sentiment of the reviews and classify the sentence into 5 or 3 different sentiments –
●Positive
●Somewhat Positive
●Neutral
●Somewhat Negative
●Negative
●Feedback Classification
●Classify the review into different classes that gives understanding of what review is talking about.
●Delivery, Billing, Product Quality, Product Features, Technical Support, Service Support etc…

19. Customer Feedback Analysis
●How to perform text classification?
●RNNs (Recurrent Neural Networks)
●Transformer Based Networks – BERT, RoBerta, XLNet…
●Example of sentiment analysis:
○“The laptop is very good to handle. It boots up pretty fast, it has very good design”.
■Positive
○“The delivery was very late. The technical support guy couldn’t understand the problem that I am
facing. It is a pathetic experience with Dell.
■Negative
○“The laptop features are really great. Though the battery life can be improved. Although the laptop
is satisfactory.
■Somewhat Positive

20. Customer Feedback Analysis
●RNNs
●For understanding the context within multiple
sentences, the neural networks we design should
hold the dependencies and connections within the
sentences.
●For language translation or next word prediction we
need to have previous word’s details.
●Networks should have some sort of memory and we
call them as recurrent neural networks (RNNs).

21. Customer Feedback Analysis
●RNNs
●In the previous diagram, a chunk of neural network, A, looks at some input xₜ and outputs a value hₜ. A loop
allows information to be passed from one step of the network to the next.
●A recurrent neural network can be thought of as multiple copies of the same network, each passing a message to
a successor.

22. Customer Feedback Analysis

23. Customer Feedback Analysis
●Problems with long term dependencies:
○Sentence 1: The clouds in the ______ are beautiful.
○Sentence 2: I grew up in France during some blahblahblah… I speak fluent ____.
○Sentence 3: “Macbook Air has been a great piece of device for me. It has some really cool features that
makes it one of the lightest laptops. This laptop boots up very fast and Apple has done it again. Their
devices are always the best. “

24. 40,000/-
Handson Project Links

●Twitter Sentiment Analysis:
https://www.analyticsvidhya.com/blog/2021/06/twitter-sentiment-analysis-a
-nlp-use-case-for-beginners/

●Named Entity Recognition:
https://www.analyticsvidhya.com/blog/2021/06/nlp-application-named-entit
y-recognition-ner-in-python-with-spacy/

●Creating Chatbot using python:
https://www.section.io/engineering-education/creating-chatbot-using-natu
ral-language-processing-in-python/

25. 40,000/-
Q/A

26. 40,000/-
Thank You