Scribd
Upload
11 views12 pages

NLP - Jan 2020

Natural language processing practices

Uploaded by

trina.desro
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
11 views12 pages

NLP - Jan 2020

Natural language processing practices

Uploaded by

trina.desro
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 12

NLP Project Updates

Jan 2020
• NLP Roadmap Review
• Progress on Data Importing
• Next Steps

Proprietary & Confidential 2


NLP Roadmap
• Platform integration
Phase
1:
Catego
• Automated consumption of email and SMS responses
rize
and
• Categorization reporting for Telus analysis and action
Report
(Q1)

• Send Email / SMS response to replies (Thank you for your payment.
Phase Someone from Telus will contact you about your issues etc.)
2: Auto
Reply • Feedback into model for fine tuning.
(Q2)

• Place in already Paid nurtures.


Phase
3:
Action:
• Place in a “Customer Defined” P2P nurture
Nurtur
e (Q3)

Future: Chat Proprietary & Confidential 3


NLP Roadmap – Q1 End Result Demo
2.1K Raw Emails

1.3K Classified

0.8K Noisy

Not enough content (too short, only emojis, autoreplies)


Bad encoding (html parsing error)
Unable to translate (Low confidence language detection/translation) Proprietary & Confidential 4
Progress on Data Importing

• Source: Telus/Koodo gmail boxes – Elio sends eml files in

Email batches
• Processing: extract body text and meta data, preprocess text
• Destination: SQL database

SMS
• Source: Twillio server
• Processing: extract body text and meta data, preprocess text
• Destination: SQL database

Loaded 1 Full year of data

Proprietary & Confidential 5


Proprietary & Confidential 6
Stable months

Proprietary & Confidential 7


Proprietary & Confidential 8
Proprietary & Confidential 9
Proprietary & Confidential 10
Next steps
• Enhance data preprocessing steps (convert noise to classifiable data)
• Start working on French language

• Enhance classification model accuracy


• Continuous data annotation and re-training

• Machine learning tricks to deal with skewed data

• Annotation of SMS data for model evaluation (so far focus was on
emails)

Proprietary & Confidential 11


Next steps
• Q1 production/integration steps
• Finalize data importing procedure

• Connect database tables to scheduled NLP procedures

• Develop the reporting platform


• Confirm format and frequency

Proprietary & Confidential 12

You might also like