What can Serverless AI/ML APIs tell us about Fake and Credible News?

A presentation at ServerlessDays Stockholm 2019 in October 2019 in Stockholm, Sweden by Bruno Amaro Almeida

What can Serverless AI/ML APIs tell us about fake and credible news media websites? BERLIN · HELSIN K I · LON DON · MUN ICH · OSLO · STOCK HOLM Bruno Amaro Almeida | 24 Oct 2019 Principal Architect & Technology Advisor | Futurice · TAMPERE @bruno_amaro Photo by Juliana Malta on Unsplash

FUTURE. CO-CREATED. Nordic Roots, Global Mindset PEOPLE NATIONALITIES 550+ 38 8 30% OFFICES Tampere Helsinki Oslo Stockholm YoY GROWTH Family of Companies eCommerce & Growth Hacking Berlin London Artificial Intelligence & Machine Learning Stuttgart Munich

AI & Analytics Capabilities Data Engineering (ingest, prepare, transform, analyze) AI/ML Platform (build, train, deploy) AI/ML API’s (pre-trained models, serverless, out of the box) @bruno_amaro

AWS vs GCP vs Azure: Data Engineering / AI Ingest ETL • AWS Kinesis • AWS Glue / EMR • Google Pub/Sub • Google Dataflow / DataProc • Azure Event Hubs • Azure DataFactory / DataBricks Raw Storage • AWS S3 • Google Cloud Storage • Azure Data Lake Storage Data Warehouse • AWS Redshift • Google Cloud BigQuery • Azure SQL Data Warehouse Machine Learning • AWS SageMarker • Google Cloud Datalab Analytics / BI • AWS QuickSight • Google Cloud Data Studio • Azure ML Studio / Workbench • Power BI @bruno_amaro

AWS vs GCP vs Azure: AI/ML API’s AI/ML Service APIs AI/ML Service APIs AI/ML Service APIs • AWS Lex • Google Dialogflow • Azure Bot Service • AWS Rekognition • Google Vision API • Azure Vision • AWS Translate • Google Text-to-Speech API (ASR) • AWS Polly (TTS) • Google Speech-to-Text API • Azure Speech • Translator Speech API, Bing Speech API • AWS Transcribe (ASR) • Google Natural Language API (NPL) • AWS Textract (OCR) • Google Translation API • Azure Knowledge • AWS Comprehend (NPL) • Google Video Intelligence API • Azure Search • AWS Forecast (Time-series forecast) • Google Inference API (Time-series forecast) • Google Job Discovery • Bing News/Web/Image/Video/Custom Search • Azure Language • Google Cloud Genomics (Store and process genomes and related experiments ) Source: AWS • In preview: Speaker Recognition API, Custom Speech Service Source: Google Cloud • Language Understanding (LUIS), Bing Spell Check, Text Analytics, Translator Text API Source: Azure @bruno_amaro

News Media Websites (Fake vs Credible)

Website Metadata Extraction Methods xvfb-run (…) wkhtmltoimage (…) lynx —dump image-scraper Pressure on Theresa May to resign will ‘increase dramatically’ following extension, warns David Davis + Review set for June 21 after Macron opposes long delay + EU already talking about possibility of further extension + Britain’s EU ambassador formally accepts extension [51]ReconstructionHow Emmanuel Macron raged against Britain’s chaotic [52]Janet DaleyAny Brexit solution with Theresa May in post is impossible [53]How long does the PM have left after being forced to accept a six month (…) @bruno_amaro

Enrich website metadata with AI/ML API Google Vision API: • Text • Labels • Safe Search • Web Entities Tusk recommends EU insist on year-long Brexit extension after MPs + PM arrives in France to meet Macron after talks with Merkel + Four Cabinet ministers rebel against vote to extend Brexit + Tusk: ‘Little reason’ to believe Brexit can be delivered by [51]ExclusiveLiam Fox tells Tory MPs that customs union will be ‘worst of [52]Daniel KawczynskiERG members are putting Brexit in jeopardy, I had no [53]Is Macron seeking his Charles de Gaulle moment over Brexit? (…) Google Natural Language API: • Analyze Sentiment • Text Classify

Enrich website metadata with AI/ML API

AWS Services • Rekognition Enrich website metadata with AI/ML API • Comprehend @bruno_amaro

Enrich website metadata with AI/ML API @bruno_amaro

AWS Comprehend for Sentiment Detection

AWS Comprehend for Sentiment Detection

AWS Comprehend for Sentiment Detection

AWS Comprehend for Sentiment Detection

Scale. Aggregate. Value? vs BERLIN · HELSIN K I · LON DON · MUN ICH · OSLO · STOCK HOLM · TAMPERE Photo by Elijah O’Donnell on Unsplash

Interesting findings Text Categories / Entities (AWS) @bruno_amaro

Interesting findings Text Categories / Entities (Google Cloud) @bruno_amaro

Interesting findings Sentiment Analysis (AWS) @bruno_amaro

Interesting findings Sentiment Analysis (Google Cloud) @bruno_amaro

Interesting findings Image Recognition & Moderation Labels (AWS) @bruno_amaro

Interesting findings Image Recognition & Safe Search Annotation (Google Cloud) @bruno_amaro

Thank you! Kiitos! Danke! Tack! Bruno Almeida P RINC IP AL ARC HITE C T & TE C HNOL OGY ADV ISOR Cloud, Security, DevOps, Data Engineering & AI Reach out on: @bruno_amaro @brunoamaroalmeida BERLIN · HELSIN K I · LON DON · MUN ICH · OSLO · STOCK HOLM · TAMPERE

Bruno Amaro Almeida
@brunoamaroalmeida

1 / 23

News media websites are - and have been in the past years - in the epicenter of multiple society debates (e.g. fake news, elections and geo-political influence, etc.) with a tremendous impact in our society. What happens when you enrich metadata collected from news media with the AI/ML API’s (e.g. Image Classification, Translation, Sentiment Analysis) that public cloud (AWS, Google Cloud and Azure) provides?

Buzz and feedback

Here’s what was said about this presentation on social media.

@bruno_amaro “What can Serverless AI/ML APIs tell us about fake and credible news media websites?” at #ServerlessSthlm pic.twitter.com/lPzvlwEGpd
— Josh Glover [He/Him] ƛ (@jmglov) October 24, 2019
Fake news on the left, credible media on the right. Fake news sites contained some images of violence and gore, credible media sites did not. pic.twitter.com/vKb3w8f89d
— Josh Glover [He/Him] ƛ (@jmglov) October 24, 2019
No, thank you! What an interesting talk!

[Slide reads: “Thank you! Kiitos! Danke! Tack!”] pic.twitter.com/5OyWQ3LwJB
— Josh Glover [He/Him] ƛ (@jmglov) October 24, 2019
It was a pleasure to speak today about #serverless AI/ML APIs at @ServerlessSthlm!
However, the highlight of my day was to facilitate a panel discussion with these incredible talented people.Warm thanks to @AngelaTimofte @GunnarGrosch @michelr79 @BretMcG @heitor_lessa Avengers🤘 pic.twitter.com/eoHaF2g7Bj
— Bruno Amaro Almeida (@bruno_amaro) October 24, 2019
Thank you for today @ServerlessSthlm and all speakers and attendees!! ❤️🥳 Here are all the smart things you said in one slide 😁😉@rchrdbyd @0x12b @AngelaTimofte @orysegal @bruno_amaro @BretMcG @heitor_lessa @Microsoft @TServerless @jbesw @MHallne @fredrik @GunnarGrosch pic.twitter.com/tfI1jSWUIJ
— David Borgenvik (@DBorgenvik) October 24, 2019
The fireside chat at #ServerlessDays in Stockholm is about to begin. It’s all about #serverless and I’m surrounded by super talented people. I'm so excited about this ❤️
#trustpilot #womenintech #aws #databases #architecture pic.twitter.com/lploGwIUkr
— Angela Timofte (@AngelaTimofte) October 24, 2019
A fireside chat with some of the #ServerlessDays speakers and organisers. pic.twitter.com/wMzGgTYbTJ
— Josh Glover [He/Him] ƛ (@jmglov) October 24, 2019
Starting off the afternoon with @bruno_amaro's session on #serverless ML and fake news. All of the big cloud providers have a lot of services available. Google is lacking a service to figure out which of its products to kill next. #ServerlessSthlm pic.twitter.com/vnDkkeDBUU
— Michael Bahr (@michabahr) October 24, 2019
"Take the results from pretrained models with a grain of salt" @bruno_amaro - #AWS thinks drugs are good? 🤔#ServerlessSthlm pic.twitter.com/gQ7Jh98wXM
— Michael Bahr (@michabahr) October 24, 2019
And we’re back!! @bruno_amaro talking about how to deal with Fake News using Serverless, AI/ML#serverlessSthlm pic.twitter.com/hvJsw6YZoZ
— Heitor Lessa (@heitor_lessa) October 24, 2019
What can Serverless AI/ML APIs tell us about fake and credible news media websites? @bruno_amaro from @futurice talks about the insights we gain when leveraging AI/ML capabilities. #ServerlessSthlm pic.twitter.com/KgNEYaoKfI
— Gunnar Grosch (@GunnarGrosch) October 24, 2019
We’re kicking off @ServerlessSthlm pic.twitter.com/Y3d3IXZ9NT
— Richard H. Boyd (@rchrdbyd) October 24, 2019