GoDaddy Customer Success Dashboard Using Apache Spark with Baburao Kamble

Copyright© 2017GoDaddyInc. AllRights Reserved.
Customer Success Dashboard
GoDaddy Business Score
Baburao Kamble; Ryan Kleck; Robert Brown

Agenda
2
GoDaddy’s Mission for Small Business Success
Machine Learning on 64M Websites Content using Spark 2.1
Business Score
Marketing Use Case

See the Internet
like never before.
hello.
Everything you
need to know to
get your business
online.

Develop the Content Analytics Platform (online and offline) for
the GoDaddy’s Customers to Understand Digital Presence of the
Small Business in a particular industry segment.

HTML
Machine Logs
Security Logs
Internal Data
Open Source Data
TE L
Communication
Score Engine
Industry Verticals
Traffic Forecast (global,local)
EconomicFootprints
Business Score
GD SBI
Go
Initiative
Marketing
Ad hoc
Analytics
User Interface
FeatureDB

Components
9
Using Customer’s website content, develop a Customer
Success Dashboard from digital presence & local
econometrics in the city/neighborhood where business is
situated .
• Customer Success Dashboard will have all the analytics of
the website done in real-time and batch according to the use
case.
• Batch Processing: We are using GD data crawler as primary
source of the data third party social & business API’s. Use
case: Customers Online Presence Index as Business
Score.
• Real-time Processing: Scrape the content of the website
and plug DNS address for various internal tools to create
features in real-time. Use case: What you should add in
your website to make small business successful.
SEO
site content &
structure
ECOM
small business
websites
SOCIAL
social media
coverage
ECON
local economy
index
WEB
website
perfection
HOSTING
faster traffic for
your contents
ECOM
small business
websites
INDUSTRY
industry profiles
by objectives

Industry Segments/Profiles
10
Using Customer’s website content, develop a Customer
Success Dashboard from digital presence & local
econometrics in the city/neighborhood where business is
situated .
• Customer Success Dashboard will have all the analytics of
the website done in real-time and batch according to the
use case.
• Batch Processing: We are using GD data crawler as
primary source of the data third party social & business
API’s. Use case: Customers Online Presence Index as
Business Score.
• Real-time Processing: Scrape the content of the website
and plug DNS address for various internal tools to create
features in real-time. Use case: What you should add in
your website to make small business successful.
TLDs
nlp on TLDs
keywords
PARTNERSHIPs
sic industry
codes
SURVEY
self
identification
CONTENT
naics
classification

TLD’s Verticalization
11
Business Need
• Know the domains and purpose of the domain in detail by natural language processing (NLP).
• Are they personal/blog/portfolio or are they business/LLC/consultancy.
• Classify domains into 6 classes identified by modelers and marketers.
• Enable domain specific solution to enable hidden business.
Model
• We have used internal tokenizer to tokenize domains
• These tokens feed to feature generation model which detect language, human names, & geography.
• Machine Learning model is trained on the features and keywords to classify the tld’s into 6 classes.
• In 2nd
iteration, business keyword rule model is designed to overwrite some misclassified data.

TLD’s Verticalization
12
containsenglish domainname hasname isallenglish keywords language pos wordcount
TRUE kidsmedicinestickers.com FALSE TRUE kids medicine stickers english N|V-N|V-N 3
TRUE actionwindowtint.com FALSE TRUE action window tint english N|V-N-N|V 3
TRUE aubihardware.com TRUE FALSE aubi hardware english N-N 2
TRUE michiganfellowship.com TRUE TRUE michigan fellowship english N-N 2
FALSE devinlamontagne.com TRUE FALSE devin lamontagne unknown V-N 2
We use Python internal tokenize and NLTK to generate the features.
Above features and keywords have used to classify domains in 6 classes based on the topics modeling some
of the classes are overwritten by business rules.
domainname Segment
kidsmedicinestickers.com Business
actionwindowtint.com Business
aubihardware.com Business
michiganfellowship.com Government
devinlamontagne.com Personal

Industry Verticalization
13
Business Need
• Identify the North American Industry Classification System (NAICS) based industry verticals for the
GoDaddy domains (64M) based on the content analysis and domain names.
• Rank each domain into industry vertical based on probabilities.
• Enable business specific marketing solution.
NAICS
• A numerical, hierarchical classification system that groups all business establishments into industries based
on production process.
• The foundation on which economic census data is collected, tabulated, analyzed, and disseminated.
• A common language used among all North America.

LDA: Latent Dirichlet Allocation
14
• We have developed training data using business keywords, NAICS description and keywords from the US
Census data.
• We used a technique called Latent Dirichlet Allocation to classify the fingerprint of the “unknown” domain to
the lineup of known NAICS Sector fingerprints.
• LDAis a generative statistical model that allows sets of observations in a linguistic corpus to be explained by unobserved groups.
• Among these algorithms, Latent DirichletAllocation (LDA), technique based in Bayesian Modeling, is the most commonly used
techniques for nowadays.
• Assumptions on all variables:
• Word: the basic unit of discrete data
• Document: a collection of words (exchangeability assumption)
• Corpus: a collection of documents
• Topic (hidden): a distribution over words & the number of topics 𝑲 is known.
LDA

𝛼
𝜃%
𝑍%,(
𝑊%,(
𝛽+
𝜂
𝑁
𝐷
𝑘
Dirichlet parameter
Per document topic
proportion
Per word topic
assignment
Observed word
Topics
Topics
hyperparameter
For each topic 𝑘,draw a multinomial over words 𝛽+~𝐷𝑖𝑟 𝜂
For each document 𝑑,
• Draw a document topic proportion 𝜽%~𝐷𝑖𝑟 𝛼
• For each word 𝑤%,(:
• Draw a topic 𝑧%,(~𝑀𝑢𝑙𝑡𝑖 𝜽%
• Draw a word 𝑤%,(~𝑀𝑢𝑙𝑡𝑖(𝛽<=,>
)
Mathematical Representation of LDA

Modeling Approach
16
Input
testing
Spark
DataFrame
ML Model
Input
training
Spark
DataFrame
Preprocessing
Preprocessing
Feature Selection &
Transformation
Feature Selection &
Transformation
Score

Copyright© 2017GoDaddyInc. AllRights Reserved.17
{NAICS} hdfs://html
Tokenize NAICS
Keywords
html2text
to tokenize
Industries to topics Websites to topics
Websites to Industries
toAnalytics
LDA
Input
Preprocessing
& NLP
Machine Learning
Output
Analytics
Industry Verticalization

Sample Results
18
Industry Vertical Sample Score for GoDaddy.com
{‘godaddy.com’:[0.0,0.037261519371214455, 0.12163335800668636, 0.12212609505637625,0.0, 0.0, 0.0391883272323731, 0.006822915544477993,
0.07649356215522182, 0.03418770216006102, 0.08399637531579494, 0.0, 0.044632956375623346,0.08008591676796364, 0.23397736516285688,
0.023100845006343924, 0.05476212126480185, 0.045210897428489265, 0.01195930071056271,0.05362539333191509, 0.030935349109237273,
0.03476925258258562, 0.028247390216576837, 0.04214249732233799,0.11761814220985195, 0.0, 0.03485210250149108, 0.0, 0.036624106810977664,
0.023490911619049086,0.08161086708509405, 0.09972968094058156, 0.024987767078109006,0.01669645821975547, 0.027306928087764512, 0.0, 0.0,
0.59144987630468255, 0.01714920401645141, 0.057044712502075504,0.33934270898168075,0.026937393520935108]
Top three Industryverticals with probabilities
Row(domainname=u'godaddy.com, 541=u' 0.591449876', 551=u' 0.339342709', 518=u’0.233977365')
518 - Data processing, hosting, and related services
541- Professional, scientific,and technical services
551- Management of Companies and Enterprise

WEB
19
• For Small business's website, it's vital
to position business online with a
strong, professional destination that
gives customers the impression you
mean business and the motivation to
want to engage more with your
business.
• We have created the process to
analyze the website based on the What
makes the Small Business Website
perfect. This helps visitor connect and
checkout the importantsection of the
website easily from website and
mobile.
WEBSITE
website
completeness
DESIGN
modern UI/UX
features templets
METEDATA
Analyze metadata
completeness.
CODING
coding style &
text to code ratio.

ECOM
20
• For online small ecommerce business,
it's vital to make the secure and
professional website to sale products
online.
• Based on prominentecommerce
websites, we have created the
important componentfor successful
ecommerce business and checked
these components on the websites.
analyze the website for the ecommerce
components (shopping cart, payment
solution,SSL, shipping,etc.) .
SECURTY
web security &
SSL encryption
CONATCT
contact, chat,
email, maps etc.
ANALYTICS
analytics plugins
& campaign.
CART
online shopping &
payment solution

HOSTING
21
• Most small businesses end up making
a wrong choice in the beginning
because they do not properly evaluate
their needs.
analyze the hosting based on the
content, SEO and website content size,
speed and hosting information
A-RECORDS
hosting
information
SPEED&COUNTS
download speed &
dns count month
BANDWITH
self
identification
CONTENT
type & size of the
content

SEO
22
• Small business’s definitely needs to
have an SEO strategy in place if you
are interested in succeeding in terms of
online marketing.
• investing in organic SEO is more
important for small businesses now
than ever before.
• We have created the process just like
SEO experts to analyze the keywords
of website’s meta title, meta
description,h1, h2, and image alt tag.
This helps search engines connectwith
website with the keyword, making web
page more relevant.
METADATA
metadata
information
LINKS
analyze all urls &
links on website
ROBOTS
check robots.txt
& sitemap
WEB
web elements &
keywords

ECON
23
• Small business’s growth is totally
depends on the local economy and
purchasing power of the town.
• Local economy has big impacton the
Small Businesses and vice a versa.
• We have used online open dataset like
tax, income,housing prices to create
the ECON index.
INCOME
average income of
the family
CITY DYNAMICS
population,
growth, tourism
HOUSING
housing price
index
TAX
local small
business tax

SOCIAL
24
• For Small business's there is need
attract customers to local shops or
online ecommerce business with a
strong online presence.
• Traffic to the website or local store can
be driven through the large social
media efforts and listing in review
websites like yelp,foursquare etc.
• We are using third party api, data
providers and listings to aggregate the
score as the index of popularity of
business.
SOCIAL
social media for
small business.
LINKS
Analyze all urls &
links on website
LISTINGS
local business
listing (e.g. yelp)
MEDIA
Web elements &
keywords

GoDaddy Business Score
• GoDaddy Small Business Index (score) is a digital number derived from a mathematical interpretation
through complex machine learning process on the digital information derived from the complex data.
• Our Score is a measure of how well website is presented, secured, socialized and how well website is doing
relative to other sites in same industry segment.
• We use the Bayesian Network to select the features and Gaussian constants from the feature database.

Web Features Relevancy by Industry Segments
Bayesian network diagram shows the simple
relationship between the different components (data) of
the Customer Success Dashboard.
For example one can see that SEO, industry, social
(Facebook, twitter, etc.), mobile optimized and site
completeness all have casual effects on the likelihood of
the website traffic as expected. The relationship between
the industry and traffic also becomes clearer.

Business Score
15%
WEB
10%
Social
21%
ECOM
16%
HOSTING
18%
ECON
Business
Score
is to based on the Credit Scoring System.
Each Component has given weight based
on the importance to the Small Business
27
20%
SEO

GoDaddy Small Business Index
GoDaddy Business Score/Index is the similar to the
credit scoring system for digital presence.
Each of digital component has given weight.
We analyze and rescale the score based on the features of
the website with top 100 domains in the each industry
segment.
GD-Business
Score Model
based on Gaussian
formula
Get the top performing domains
in each Vertical based on the
Bayesian Network of Feature
Data
FeatureDB
Feature (120) database
created from Spark ETL
data pipelines
{‘domainname’: ‘godaddy.com’,
‘518’: ’98.8’,
‘541’: ’93.2’,
‘551’: ’97.3’}

Marketing Use Cases
29

Mobile Friendly
Websites
30
• Let's make sure GoDaddy
Customer's website shows
up in search results.
• Detect the whether
website is configured for
multiple devices & help
search engines understand
Small Business website to
show in mobile device
based search.
• We have designed the
algorithm to detect the
responsive web design
which is recommended
design pattern by major
search engines.
• Algorithm also detect
mobile optimized
websites.
• Algorithm is validated
with Google’s Mobile
Friendly tool with 0.2%
sample which reported
94% accuracy.
• We send notification to the
shopper and advice them
based on the outcome of the
mobile friendliness of the
websites.
• GoDaddy also help
shoppers to make websites
mobile friendly on all
devices.
Increase Mobile
device based SEO
Text Analytics on Raw
HTML data
Help to Small
Business.
MISSION MODEL REVENUE

Cart + Payment
- SSL
31
• Let's make sure GoDaddy
Customer's online small
businesses are safely
accept the credit cards.
• Detect the whether
website has shopping cart,
payment solution and
SSL.
• We have designed the
algorithm to detect the
SSL, it’s type, shopping
cart and payment solution
provider.
• Algorithm also SSLdetect
encryption type.
• We send advice shoppers
based on the outcome of the
features detected from the
websites.
• GoDaddy also help
shoppers to make websites
more secure and shopper
friendly to accept online
transactions.
Secure the
transactions
Text Analytics on Raw
HTML data
Help to Ecommerce
Small Business.
MISSION MODEL REVENUE

GoDaddy’sBusiness Score goal is to help
established small businesses get to the next level
by increasing online presence.
32

GoDaddy Customer Success Dashboard Using Apache Spark with Baburao Kamble

More Related Content

What's hot (14)

Similar to GoDaddy Customer Success Dashboard Using Apache Spark with Baburao Kamble (20)

More from Databricks (20)

Recently uploaded (20)

GoDaddy Customer Success Dashboard Using Apache Spark with Baburao Kamble