Pranay Mathur and Aman Gill
Data Science Interns
Edelweiss Springboard: Projects in Big data
and Machine Learning
COGNITIVE EMAIL AUTO TAGGER
2
Problem Statement : To categorize incoming customer e-mails.
Why is this needed ?
 To reduce man hours spent on
categorizing e-mails.
 To reduce the possibility of late
responses to customer needs.
Objective
To automate the e-mail categorization
task.
To improve Company-Customer
relations by quickening responses to
customer.
We are providing
 Automatic responses to multiple
requests in an email.
Self learning model - Will improve
with time.
System to increase quick responses
to customer, hence improving
customer relations.
Way Forward
 Can be used for other Edelweiss
business units.
 As data for minority category
increases, they can be predicted as
well.
 Multi-Label Categorization as data
improves.
COGNITIVE EMAIL AUTO TAGGER
After AutomationBefore Automation
Data Flow
Approach
Training Module
Previous Emails
Categories
Trained Model
Email Auto -Tagger
Uncategorized
New Emails
Trained Model
 Auto-Assigned Category.
 Other Suggested Categories
 Actionable Dates
 Related Loan No.s
 Instant pulse response for STPs*
*STP: Straight through Process. Processes which can be dealt automatically.
Example: Statement of Accounts, IT Certificate for previous financial years
Feedback
Results
5
Status and Result
Categorization model implemented with accuracy of 85% ± 3
Captures Straight Through Process (Automated responses) with high accuracy
 Integration with the live server ongoing
 Gives better result than manual categorization done by customer representatives
Implemented.
Approved by business.
OCR Driven Creditor/Debtors PDF to Excel
6
Problem Statement : To convert Creditors/Debtors Pdf files to Excel
Why is this needed ?
 To reduce man hours spent on
manually typing the table data into
excel.
 A sub-module for an on-going
marketing project in Edelweiss.
Objective
To convert the Pdf files into de-
skewed images.
To extract tables from these images
by developing a user-friendly
interface.
We are providing
 Pdf to image conversion.
 GUI for cropping area of interest.
 Extracting table data and structure
from images using OCR with very
good accuracy
Way Forward
Researching profiles of companies
present in Creditors/Debtors table in
business with Edelweiss
 Finding customers who might require
a loan with 2% hit rate compared to
regular advertising strategies(0.4%).
OCR Driven Creditor/Debtors PDF to Excel
7
Data Flow
1.Converts PDF to
Image.
2.De-Skews Image
CROP Area of Interest
OCR
GUI Module
Creditors, Debtors
PDFs
Implemented.
Used by business.
8

64b72250-4845-4df6-bdf8-5a2393ade962-160920133855

  • 1.
    Pranay Mathur andAman Gill Data Science Interns Edelweiss Springboard: Projects in Big data and Machine Learning
  • 2.
    COGNITIVE EMAIL AUTOTAGGER 2 Problem Statement : To categorize incoming customer e-mails. Why is this needed ?  To reduce man hours spent on categorizing e-mails.  To reduce the possibility of late responses to customer needs. Objective To automate the e-mail categorization task. To improve Company-Customer relations by quickening responses to customer. We are providing  Automatic responses to multiple requests in an email. Self learning model - Will improve with time. System to increase quick responses to customer, hence improving customer relations. Way Forward  Can be used for other Edelweiss business units.  As data for minority category increases, they can be predicted as well.  Multi-Label Categorization as data improves.
  • 3.
    COGNITIVE EMAIL AUTOTAGGER After AutomationBefore Automation Data Flow
  • 4.
    Approach Training Module Previous Emails Categories TrainedModel Email Auto -Tagger Uncategorized New Emails Trained Model  Auto-Assigned Category.  Other Suggested Categories  Actionable Dates  Related Loan No.s  Instant pulse response for STPs* *STP: Straight through Process. Processes which can be dealt automatically. Example: Statement of Accounts, IT Certificate for previous financial years Feedback
  • 5.
    Results 5 Status and Result Categorizationmodel implemented with accuracy of 85% ± 3 Captures Straight Through Process (Automated responses) with high accuracy  Integration with the live server ongoing  Gives better result than manual categorization done by customer representatives Implemented. Approved by business.
  • 6.
    OCR Driven Creditor/DebtorsPDF to Excel 6 Problem Statement : To convert Creditors/Debtors Pdf files to Excel Why is this needed ?  To reduce man hours spent on manually typing the table data into excel.  A sub-module for an on-going marketing project in Edelweiss. Objective To convert the Pdf files into de- skewed images. To extract tables from these images by developing a user-friendly interface. We are providing  Pdf to image conversion.  GUI for cropping area of interest.  Extracting table data and structure from images using OCR with very good accuracy Way Forward Researching profiles of companies present in Creditors/Debtors table in business with Edelweiss  Finding customers who might require a loan with 2% hit rate compared to regular advertising strategies(0.4%).
  • 7.
    OCR Driven Creditor/DebtorsPDF to Excel 7 Data Flow 1.Converts PDF to Image. 2.De-Skews Image CROP Area of Interest OCR GUI Module Creditors, Debtors PDFs Implemented. Used by business.
  • 8.