Jos Polfliet
Specialist Data Sciences at SAS Institute - @JosPolfliet
Data Innovation Summit
March, 30 2017
#DIS2017
Suicide prevention using text
analytics
can someone with my number please text me because i have nobody to
talk to and i just want to hut myself so badly and i’m scared
my own father told me I deserved being killed. I hate him
there's not one day I don’t think of killing myself
Source
Statistics Canada
Canadian Mental Health Association
1 out of 4 teenager deaths
2nd leading cause of death
1.1M tweets
Feelings: 106k ≈ 10%
Bullying: 11k ≈ 1%
Suicide: 382 ≈ 0.035% of tweets
39% correlation between
suicide and bullying
Data Impact Challenge
Question 3
What proportion of Canadian youth (13-17) post about their mental health, and
describe experiencing bullying or suicidal thoughts in the past 12 months on social
media?
Our submission
Won $10,000
donated to mental health charities
mind your mind and Rise Asset Development
Canada Health Infoway
$95,000 in awards
Download Language Age Topic Analyze
Tweets
in Canada
during a specific
time frame
Detect language
Filter English
Build predictive
model for Age
Filter teenagers
Detect topic
• Aggression
• Alcohol
• Bullying
• Family
• Feelings
• Relationships
• …
Analyze
relationships
between topics
Contextual
analysis of topics
Python crawler for
Twitter API
SAS Text Analytics SAS Enterprise Miner
SAS Text Analytics
SAS Text Analytics SAS Visual Analytics
How can we predict “At-Risk”
social media users (for
suicide and self-harm)?
Data Science Project Funnel
Idea Define Prove Decide Build Deploy
Data available?
Does this make sense?
Done before?
Input variables?
Target variable?
Expected value?
Is performance as we
expect?
Data Quality?
Offline evaluation
Potential value?
Accuracy measures?
How difficult?
Feasibility?
Data sources
(internal/external)?
Time?
Priority?
Budget?
Executive buy-in?
Business process
redesign?
Architecture
Technical
Performance
Change mgmt.
Resources
Actual value?
Live Evaluation
WE ARE HERE
Use your textual data.
Call Center Notes Survey Feedback
Online Forums Blogs Consumer Reviews Online News Social Networks
Associate Comments Claims & Case NotesResearch & Publications
Live Chat Factory/Technician Notes HR data Medical/Health Records Contracts & Applications
If anybody knows Maggie, let me know! [I’m serious]
@JosPolfliet
Jos Polfliet
Specialist Data Sciences at SAS Institute - @JosPolfliet
Data Innovation Summit
March, 30 2017
#DIS2017
Suicide prevention using text
analytics

Suicide prevention using social media analytics

  • 1.
    Jos Polfliet Specialist DataSciences at SAS Institute - @JosPolfliet Data Innovation Summit March, 30 2017 #DIS2017 Suicide prevention using text analytics
  • 2.
    can someone withmy number please text me because i have nobody to talk to and i just want to hut myself so badly and i’m scared my own father told me I deserved being killed. I hate him there's not one day I don’t think of killing myself
  • 3.
    Source Statistics Canada Canadian MentalHealth Association 1 out of 4 teenager deaths 2nd leading cause of death
  • 4.
    1.1M tweets Feelings: 106k≈ 10% Bullying: 11k ≈ 1% Suicide: 382 ≈ 0.035% of tweets 39% correlation between suicide and bullying
  • 5.
    Data Impact Challenge Question3 What proportion of Canadian youth (13-17) post about their mental health, and describe experiencing bullying or suicidal thoughts in the past 12 months on social media? Our submission Won $10,000 donated to mental health charities mind your mind and Rise Asset Development Canada Health Infoway $95,000 in awards
  • 6.
    Download Language AgeTopic Analyze Tweets in Canada during a specific time frame Detect language Filter English Build predictive model for Age Filter teenagers Detect topic • Aggression • Alcohol • Bullying • Family • Feelings • Relationships • … Analyze relationships between topics Contextual analysis of topics Python crawler for Twitter API SAS Text Analytics SAS Enterprise Miner SAS Text Analytics SAS Text Analytics SAS Visual Analytics
  • 7.
    How can wepredict “At-Risk” social media users (for suicide and self-harm)?
  • 10.
    Data Science ProjectFunnel Idea Define Prove Decide Build Deploy Data available? Does this make sense? Done before? Input variables? Target variable? Expected value? Is performance as we expect? Data Quality? Offline evaluation Potential value? Accuracy measures? How difficult? Feasibility? Data sources (internal/external)? Time? Priority? Budget? Executive buy-in? Business process redesign? Architecture Technical Performance Change mgmt. Resources Actual value? Live Evaluation WE ARE HERE
  • 11.
  • 12.
    Call Center NotesSurvey Feedback Online Forums Blogs Consumer Reviews Online News Social Networks Associate Comments Claims & Case NotesResearch & Publications Live Chat Factory/Technician Notes HR data Medical/Health Records Contracts & Applications
  • 13.
    If anybody knowsMaggie, let me know! [I’m serious] @JosPolfliet
  • 14.
    Jos Polfliet Specialist DataSciences at SAS Institute - @JosPolfliet Data Innovation Summit March, 30 2017 #DIS2017 Suicide prevention using text analytics