Turning Analysis into Action with APIs - Superweek2017

Post on 18-Feb-2017

336 views 2 download

Transcript of Turning Analysis into Action with APIs - Superweek2017

TURNING ANALYSIS INTO ACTION WITH APIs

• Digital business focus• 15+ years of experience• Web Analytics House

2013, 2014, 2015, 2016, 2017

• Value-driven business• More than 40 people in

Denmark, Sweden & Norway

About IIH Nordic

About Us

Mark Edmondson• In digital since 2006• Data Insight

Developer• Google Developer

Expert for Google Analytics

• RStudio Advocate• cloudyR contributor• @HoloMarkeD

Peter Meyer• In digital since 1997• Analytics Technician• TMS and Analytics

implementer• @pmeyerdk

Agenda

• What are actionable analysis APIs?

• Example 1 - Weather API - IBM

• Example 2 - Custom prediction API - Prefetching

• Example 3 - Text sentiment API – Google/Algorithmia Machine Learning

• Future opportunities

What are Analysis

APIs?

Collect Data

Analyse Data

Change something

An Analytics Workflow

Collect Data

Analyse Data

Change somethin

g

Reporting APIs

Tracking APIs

Analysis APIs

An Analytics Workflow

Collect Data

Analyse Data

Change somethin

g

Reporting APIs

Tracking APIs

Analysis APIs

An Analytics Workflow

Analysis API

An API that transforms data into useful information

Examples of actionable analysis APIs

• Visualisations for decision support (dashboards)

• Multivariant testing• Bid managers

• Enrichment of data - weather• Content prediction - prefetching• Text analysis – machine learning API

We will talk today about these ones:

…but others include:

Weather API(data

enrichment)

Deployment• Only one webservice used• Geo- and weather data in one go• A bunch of information returned

Deployment

Analysis• Take daily lag to

negate spurious trend correlation

• tscount() R package for poisson distribution

• 16% ±4% extra visitors on rainy days

URL Prefetching

(react to predictions)

Deployment

http://code.markedmondson.me/predictClickOpenCPU/supercharge.html

https://www.noisetosignal.io/2016/11/using-google-analytics-to-predict-clicks-and-speed-up-your-website/

Simpler method using offline analysis and deploying through a GTM lookup table

Text Sentiment

API (surfacing important

features)

Real-time sentiment tracking of customer service forums/comment sections

Categorise topics your customers care about

http://bit.ly/Superweek2017-Demo

Deployment - Algorithmia• Stanford NLP Sentiment Analysis• https://algorithmia.com/

Deployment - AlgorithmiaI really love ice cream very much!

Deployment - Google• Google Cloud Natural Language API (beta)• https://cloud.google.com/natural-language/

Deployment - GoogleI really love ice cream very much!

http://glaforge.appspot.com/article/sentiment-analysis-on-tweets

Recent example at gov.uk

https://gdsdata.blog.gov.uk/2016/12/20/using-machine-learning-to-classify-user-comments-on-gov-uk/

Future Opportunitie

sThe Horizon

Machine Learning APIs

Preparation for the singularity

• Get your data tidy and out of silos• Start prototyping models• Skill up in a ML tech stack

Skill up in a machine learning tech stack

• Google BigQuery

• Google Compute Engine

• Tensorflow • Dataproc

(Spark)

• Redshift/Athena

• Amazon EMR• EC2• S3• Amazon

Machine Learning

• SQL Data Warehouse

• Azure• Microsoft R• Microsoft

Cognitive

Programming languages: R, Python, bash, JavaScript, Scala, Java, Julia(?), SQL

ML: Google FlavouredBuild your own self-driving car!

....or for simpler use cases use the pre-trained models available using easy to use APIs

A Microsoft Cognitive API Example

https://matt-stannard.blogspot.dk/2016/12/measuring-footfall-with-google.html

• Look to reduce time to actions via APIs• GTM can do more than tracking• Skill up in a machine learning tech stack

Summary

Thank you!Mark Edmondsonmark@iihnordic.com

@HoloMarkeD

https://www.linkedin.com/in/markpeteredmondson

Peter Meyerpeter.meyer@iihnordic.com

@pmeyerdk

https://www.linkedin.com/in/pmeyerdk