Intelligent Stream Mining and Strategic Response Solutions “ Real-time Internet conversations occur on many platforms from blogs to social media sites and new web superstars like Twitter. Tools that you have used in the past, like Google or their email alerts, cannot keep pace with these conversations.”
Our Core Strategy Build Web Apps that look into the Data Mine: Live Data Mine of Conversations and Mentions Social Search Engagement Work Flow Real Time & Historical Analysis
PeopleBrowsr Clients
Advertising Stream Print Radio TV Social Media Media Spend $$$ Social Media Is an efficient  channel
Inside PeopleBrowsr Analytics Fan Pages Mentions, RTs, … Comments, Sharing,… Profiles, Comments, … Status Updates, Comments, … Pictures, Comments, … Connections, Comments
Inside PeopleBrowsr  360 days of data 1 TB of data/month and growing FILTER  Tweets, BIOs, Location Full Feed from Twitter with greater API access for large scale campaigns
Problems we Solve: How do Traditional Channels effect Social Media Viral Sentiment and Customer Service Issues How do we influence the Conversation Filter High Velocity Streams for Actionable  Messages How to build an engaged Audience Converting Social Media into Sales Calculating ROI
Media Spend on Twitter Media Spend on Twitter
The PeopleBrowsr Cycle Historical and Ongoing Sentiment Analysis Campaigns to Build Followers Whitelist and Campaigns to Direct Message Reporting on ROI for traditional and Social Media Custom Monitoring Dashboards
SuperBowl Analytics:  Effect of Traditional Media  on Social Media Mechanical Turk to measure a ccurate  Sentiment Metrics to measure Success: Total Mentions Positive Mentions By Volume Mullen and Radian6
SuperBowl Results:  103,158 Total Mentions Sampled 1000 Tweets from Every Brand and used Mechanical Turk Human Sentiment to analyse Polarised: 50% Positive 28% Neg 18% Neutral
SuperBowl Positive Sentiment Chart
SuperBowl Positive Sentiment Chart
SuperBowl
SuperBowl
SuperBowl
SuperBowl
SuperBowl
O’Reilly and PeopleBrowsr Analysis:
Correlation of Tweets and Ads
Doritos and Bud Light ran a series of themed ads Doritos ran more ads and received more positive sentiment tweets
Top Brands – By Sentiment Google and Snickers had most positive sentiment Both show a longer tail of tweet interest
Low Sentiment Brands Pre-game controversy for Focus on the Family ad Fewer tweets than other brands in study GoDaddy ran racy ads
No Sentiment Brands Coca Cola and Budweiser generated more neutral sentiment than other brands
Korean Cars Kia generated more tweet volume while running fewer ads
SuperBowl By Volume Mullen and Radian6 O’Reilly and PeopleBrowsr O’Reilly and PeopleBrowsr By Volume By Positive http://www.slideshare.net/peoplebrowsr/superbowl-3231030 Contact me for an xls of the Brands Tweets  [email_address]
Hollywood Company Size:  $200 Million in Revenue Social Media Spend:  $20K Goal:  Evaluate impact of non-Traditional Media on the Social Media sphere
Hollywood Solution:  180 day Historical Analysis of Posts overlayed on TV Ad spend and other channels Performance:  Identified key performer as branded video content 50% fluctuation on engagement based on time of message release Targeted influencers RTed a combined 879 times
Hollywood  Sentiment Campaign Listen to your Audience:  Monitor Twitter for mentions of the movie and its stars Find all negative comments about the movie Order your results by influence Build Relevant Followers:  Follow most relevant people Wait for follow back Make Contact:  Acknowledge – Let complainers know they have been heard Engage with the influencers Transform enemies in your best advocates:  Run positive sentiment campaigns
Filtering a high Velocity  Stream Seek an effective way to measure brand sentiment accurately.  The goal is to find a list of influencers speaking in both positive and negative terms and engage. Risks:  Stream includes spam, affiliates, and other non-relevant mentions .  Velocity  10,000 Mentions/day
  Filtering Process Build a unique @name list from the extract Build spam @names with PB spam alogorithm Build affiliates list Review Spam and Affiliate @names - reinsert false positives  Finalise Spam and Affiliate list Wash spam and Affiliate list against the full stream to produce a non-Spam and Non-Affiliate cleaned Daily Stream Data Mine | Analytics |  Brand Engagement  |   Messaging
Sentiment Analysis 95%  accuracy Vs 70-80%  automation alone   SENTIMENT One  Build a live stream Two  Analyze with a special dictionary Three  I ncrease quality with  Real-time human re-sorting Mechanical Turk HISTORICAL ARCHIVE One  Identify Keyword @name Two  Set Time Length (30d) Three  Search archive (360d)
one  we build a live stream for an industry or a group using @Names or Keywords. Eg: Foo Camp attendees and the companies that they represent or Airlines. How it works
How it works two  we then analyze that stream with a special dictionary for the industry or group and sort the results into buckets.
How it works three  increase quality with  real time human re-sorting
Eg: Quality Improvement on June Airlines data
Custom Dashboards and Reporting
Social Media Metrics for ROI Metrics on “Time for Engagement” Best Time for Engagement with Followers
Sneak Preview: T2 – Next Gen Combine Search and Posting Provide inline Content Contextual Ads as a Post is created HyperConnected HyperLocal Integrated with other Services  Yelp Amazon Open Table
T2 prototype
T2 prototype
Intelligent Stream Mining and Strategic Response Solutions @WingDude  |  [email_address]

Gravity Summit 2010 PeopleBrowsr

  • 1.
    Intelligent Stream Miningand Strategic Response Solutions “ Real-time Internet conversations occur on many platforms from blogs to social media sites and new web superstars like Twitter. Tools that you have used in the past, like Google or their email alerts, cannot keep pace with these conversations.”
  • 2.
    Our Core StrategyBuild Web Apps that look into the Data Mine: Live Data Mine of Conversations and Mentions Social Search Engagement Work Flow Real Time & Historical Analysis
  • 3.
  • 4.
    Advertising Stream PrintRadio TV Social Media Media Spend $$$ Social Media Is an efficient channel
  • 5.
    Inside PeopleBrowsr AnalyticsFan Pages Mentions, RTs, … Comments, Sharing,… Profiles, Comments, … Status Updates, Comments, … Pictures, Comments, … Connections, Comments
  • 6.
    Inside PeopleBrowsr 360 days of data 1 TB of data/month and growing FILTER Tweets, BIOs, Location Full Feed from Twitter with greater API access for large scale campaigns
  • 7.
    Problems we Solve:How do Traditional Channels effect Social Media Viral Sentiment and Customer Service Issues How do we influence the Conversation Filter High Velocity Streams for Actionable Messages How to build an engaged Audience Converting Social Media into Sales Calculating ROI
  • 8.
    Media Spend onTwitter Media Spend on Twitter
  • 9.
    The PeopleBrowsr CycleHistorical and Ongoing Sentiment Analysis Campaigns to Build Followers Whitelist and Campaigns to Direct Message Reporting on ROI for traditional and Social Media Custom Monitoring Dashboards
  • 10.
    SuperBowl Analytics: Effect of Traditional Media on Social Media Mechanical Turk to measure a ccurate Sentiment Metrics to measure Success: Total Mentions Positive Mentions By Volume Mullen and Radian6
  • 11.
    SuperBowl Results: 103,158 Total Mentions Sampled 1000 Tweets from Every Brand and used Mechanical Turk Human Sentiment to analyse Polarised: 50% Positive 28% Neg 18% Neutral
  • 12.
  • 13.
  • 14.
  • 15.
  • 16.
  • 17.
  • 18.
  • 19.
  • 20.
  • 21.
    Doritos and BudLight ran a series of themed ads Doritos ran more ads and received more positive sentiment tweets
  • 22.
    Top Brands –By Sentiment Google and Snickers had most positive sentiment Both show a longer tail of tweet interest
  • 23.
    Low Sentiment BrandsPre-game controversy for Focus on the Family ad Fewer tweets than other brands in study GoDaddy ran racy ads
  • 24.
    No Sentiment BrandsCoca Cola and Budweiser generated more neutral sentiment than other brands
  • 25.
    Korean Cars Kiagenerated more tweet volume while running fewer ads
  • 26.
    SuperBowl By VolumeMullen and Radian6 O’Reilly and PeopleBrowsr O’Reilly and PeopleBrowsr By Volume By Positive http://www.slideshare.net/peoplebrowsr/superbowl-3231030 Contact me for an xls of the Brands Tweets [email_address]
  • 27.
    Hollywood Company Size: $200 Million in Revenue Social Media Spend: $20K Goal: Evaluate impact of non-Traditional Media on the Social Media sphere
  • 28.
    Hollywood Solution: 180 day Historical Analysis of Posts overlayed on TV Ad spend and other channels Performance: Identified key performer as branded video content 50% fluctuation on engagement based on time of message release Targeted influencers RTed a combined 879 times
  • 29.
    Hollywood SentimentCampaign Listen to your Audience: Monitor Twitter for mentions of the movie and its stars Find all negative comments about the movie Order your results by influence Build Relevant Followers: Follow most relevant people Wait for follow back Make Contact: Acknowledge – Let complainers know they have been heard Engage with the influencers Transform enemies in your best advocates: Run positive sentiment campaigns
  • 30.
    Filtering a highVelocity Stream Seek an effective way to measure brand sentiment accurately. The goal is to find a list of influencers speaking in both positive and negative terms and engage. Risks: Stream includes spam, affiliates, and other non-relevant mentions . Velocity 10,000 Mentions/day
  • 31.
    FilteringProcess Build a unique @name list from the extract Build spam @names with PB spam alogorithm Build affiliates list Review Spam and Affiliate @names - reinsert false positives Finalise Spam and Affiliate list Wash spam and Affiliate list against the full stream to produce a non-Spam and Non-Affiliate cleaned Daily Stream Data Mine | Analytics | Brand Engagement | Messaging
  • 32.
    Sentiment Analysis 95% accuracy Vs 70-80% automation alone SENTIMENT One Build a live stream Two Analyze with a special dictionary Three I ncrease quality with Real-time human re-sorting Mechanical Turk HISTORICAL ARCHIVE One Identify Keyword @name Two Set Time Length (30d) Three Search archive (360d)
  • 33.
    one webuild a live stream for an industry or a group using @Names or Keywords. Eg: Foo Camp attendees and the companies that they represent or Airlines. How it works
  • 34.
    How it workstwo we then analyze that stream with a special dictionary for the industry or group and sort the results into buckets.
  • 35.
    How it worksthree increase quality with real time human re-sorting
  • 37.
    Eg: Quality Improvementon June Airlines data
  • 38.
  • 39.
    Social Media Metricsfor ROI Metrics on “Time for Engagement” Best Time for Engagement with Followers
  • 40.
    Sneak Preview: T2– Next Gen Combine Search and Posting Provide inline Content Contextual Ads as a Post is created HyperConnected HyperLocal Integrated with other Services Yelp Amazon Open Table
  • 41.
  • 42.
  • 43.
    Intelligent Stream Miningand Strategic Response Solutions @WingDude | [email_address]

Editor's Notes

  • #22 Note higher scale for Tweet trends chart for Doritos and Bud Light (20K vs 5K for other brands)
  • #23 Note different scale from first slide (5K tweets vs 20K tweets) – far fewer tweets than Doritos and Bud Light
  • #24 Note different scale from first slide (5K tweets vs 20K tweets) – far fewer tweets than Doritos and Bud Light
  • #25 Note different scale from first slide (5K tweets vs 20K tweets) – far fewer tweets than Doritos and Bud Light
  • #26 Note different scale from first slide (5K tweets vs 20K tweets) – far fewer tweets than Doritos and Bud Light