we help you understand audience attention
play

We help you understand audience attention. Follow me: @amontalenti - PowerPoint PPT Presentation

We help you understand audience attention. Follow me: @amontalenti Website: parse.ly Our research: @parsely Blog: blog.parse.ly Our podcast: @attnpod Email: andrew@parsely.com How? Parse.ly Analytics. Web content visits represent attention at


  1. We help you understand audience attention. Follow me: @amontalenti Website: parse.ly Our research: @parsely Blog: blog.parse.ly Our podcast: @attnpod Email: andrew@parsely.com

  2. How? Parse.ly Analytics.

  3. Web content visits represent attention at global scale. Sites with content and audience Platforms + hundreds of other companies who run thousands of high-tra ff ic sites. + the long tail.

  4. Parse.ly measures content and audience … Page views Visitors Engaged time Social shares Audience loyalty Devices Video Titles Authors Sections Tags Referrers + Campaigns Publish dates Channels Much more

  5. … to tell the story behind the story.

  6. Our dashboard can answer this question: What’s gaining attention on your sites and apps? Provide a real-time and historical window into what’s happening with your content when it comes to audience attention. • 30,000 monthly active users across 350+ media companies. • Measures the attention of over 2 million page views per minute at peak time. • Sub-second data latency with 99.99% internal SLO .

  7. We make data accessible and essential.

  8. Parse.ly Analytics: 
 What’s running under the hood? Powered by mage : • 100+ Elasticsearch nodes storing over 20 terabytes of production live query data. • 3,600+ real-time processing CPU cores using Storm. • Kafka and Cassandra for rock-solid distributed streaming data . • Elastic scalability for hourly and nightly jobs using Spark.

  9. Parse.ly Analytics: 
 What does the team release publicly? We love open source! • streamparse is our publicly-maintained and popular project for running production + PyKafka parallel computation systems with Python + parsely_raw_data 2.x and 3.x, using Apache Storm. + time-engaged + others • PyKafka is the community’s fastest and most production-tested Python driver for Apache Kafka.

  10. Why now? Parse.ly Currents.

  11. Aggregate attention data already guides the industry.

  12. And answers questions it could never answer on its own .

  13. Our network data can answer this question: What do people care about? Front row seat to the web interests of over 1 billion people per month and 150 million people per day . Categories include: news, entertainment, finance, politics, sports, opinion, culture, and more. Apply modern machine learning and natural language processing techniques.

  14. Parse.ly Currents: 
 What is our petabyte-scale analysis stack? Petabytes of event data and terabytes of web crawl data. • BigQuery used with day-partitioned tables to do fast aggregation over petabyte-scale event data without running a cluster. • PyData stack used for statistics and machine learning over time series data. • Natural language processing on text data using Python, leveraging a web-based ontology (knowledge graph), domain-specific keyword/entity lists, word vectors, document classifiers, unsupervised clustering, and more.

  15. 1 billion unique visitors per month 20 billion page views per month 5 billion clicks from search, social, & others 900k posts published and analyzed each day 2 million topics, categories, and keywords

  16. Does discovery vary by topic?

  17. Lifestyle 87.1% Facebook 6.7% 6.2% Google 110k posts Local Events 61.4% 12.3% 26.2% 96k posts Entertainment 60.8% 10.1% 29.1% 190k posts U.S. Presidential Politics Topics are derived from 59.5% 15.9% 24.6% 110k posts posts in the Parse.ly Education & Research network of sites from 58.9% 19.8% 21.3% 36k posts 2016 using a topic modeling algorithm Criminal Justice 53.5% 22.2% 24.4% called LDA (Latent 55k posts Dirlichet Allocation). 
 Local Crime & Incidents For more information: 52.7% 22.6% 24.6% 98k posts parsely.com/authority National Security 41.3% 28.9% 29.7% 49k posts World Economy 36.3% 20.7% 43.0% 26k posts State & Local Politics 35.5% 22.3% 42.2% 17k posts Technology 21.3% 18.0% 60.8% 67k posts Sports 19.2% 30.4% 50.4% 210k posts Business & Finance 14.1% 39.0% 14.1% 39k posts Job Postings 11.9% 3.7% 84.4% 2.7k posts

  18. World Economy External referral sources Number of posts for each topic C O M M O N W O R D S I N P O S T S CHINA Google Search 26k posts 43.0% OIL news.google.com 4.6% PERCENT EU PER twitter.com 4.0% ENERGY World Economy SINCE Facebook TRADE yahoo! 2.4% CHINESE 36.3% EUROPEAN drudgereport.com 1.4% ACCORDING flipboard.com MARKETS 1.1% TRADING bing 0.9% BILLION linkedin.com BRITAIN 0.9% reddit.com MARKET 0.8% tra ic.outbrain.com STOCK 0.7% Device tra ic breakdown WORLD GLOBAL POWER Other STOCKS BREXIT 46% 45% 9% PRICES 20.7% DEAL BANK CENT Desktop Mobile Tablet NFL AP UK External referral sources Number of posts for each topic U.S. Presidential Politics C O M M O N W O R D S I N P O S T S Facebook 110k TRUMP posts 59.5% CLINTON news.google.com 4.3% PRESIDENT CAMPAIGN DONALD U.S. Pres. Politics twitter.com REPUBLICAN 4.1% PRESIDENTIAL ELECTION HILLARY OBAMA drudgereport.com 1.9% PARTY DEMOCRATIC Google Search CANDIDATE yahoo! 1.1% POLITICAL SANDERS bing 24.6% 0.9% WHITE HOUSE reddit.com 0.7% VOTE Device tra ic breakdown STATE COUNTRY DEBATE AMERICA WOMEN Other AMERICAN 43% 47% 10% FORMER CRUZ 15.9% NATIONAL STATES Desktop Mobile Tablet NEWS VOTERS

  19. Can Internet attention predict public opinion?

  20. Can Internet attention predict a film’s revenue?

  21. Revenue Compared to Revenue Compared to Revenue Compared to Unique Views Print Ad Cost in US $ Production Cost in US $ for Related Web Posts 3 Days Prior to Release Cumulative Box O f ice Cumulative Box O f ice Cumulative Box O f ice Gross Revenue Gross Revenue Gross Revenue 600k 600k 600k 500k 500k 500k 400k 400k 400k 300k 300k 300k 200k 200k 200k 100k 100k 100k 200k 400k 600k 800k 1M 10k 20k 30k 40k 50k 60k 70k 50k 100k 150k 200k 250k Unique Views Print Ad Cost in US $ Negative Cost in US $ 0.955 0.474 0.829 Pearson Correlation Coe f icient Pearson Correlation Coe f icient Pearson Correlation Coe f icient when excluding PG rated movies when excluding PG rated movies when excluding PG rated movies Total unique views for posts related to a Movies rated PG movie three days prior to its release has Movies not rated PG the highest correlation with revenue compared to production cost and advertising budget.

  22. Revenue Compared to Unique Views for Related Web Posts 3 Days Prior to Release 600k 500k 400k Cumulative Box 0.955 O f ice Gross 300k Pearson Correlation Coe f icient Revenue when excluding PG rated movies 200k 100k Movies rated PG 200k 400k 600k 800k 1M Movies not rated PG Unique Views

  23. We are a partner you can trust. 400+ paying clients. 3000+ big sites. 1B+ network visitors. We’re small and nimble, yet we operate with scale and integrity. We are 70+ people . • A client services, support, and ops team of 40 people , with a head o ff ice in NYC. • A fully distributed product team of engineers, data scientists, and designers. 30 people across US, Canada, and Europe. • $12M+ USD in financing raised from 2011 to 2017.

  24. Three asks for the audience today.

  25. 
 Sign up free, give us feedback! 
 http://parse.ly/currents

  26. Follow me on Twitter! @amontalenti

  27. Let’s continue the conversation about internet attention. Follow me: @amontalenti Website: parse.ly Our research: @parsely Blog: blog.parse.ly Our podcast: @attnpod Email: andrew@parsely.com

Recommend


More recommend