How To Predict Social Media Success Through Emotions
1. How To Predict Social Media Success
marketing through the power of emotions
Neuromarketing World Forum | Barcelona | 27th March 2015
2. Scale
• cameras everywhere
• basic emotions universal
• easiest to share feedback
Value
• rich data to work with
• drives thoughts and decisions
• ROI-proof building up
Why Emotions?
www.steviewonder.org.uk
3. From Push to Pull
• media has totally changed
• consumers in driver seat
• push has weaker ROI
From Reach to Relevance
• large impressions and GRPs?
• … or rather any concrete
amount of business results?
Why Social?
www.steviewonder.org.uk
4. “Half of my ad spend is
wasted but I don’t know
which half.”
John Wanamaker – US Pioneer in Marketing, 1874
7. MeasurementTargetingTesting
5% using neuro
FinishStart
< 0.1% using neuro < 0.001% using neuro
99% of current neuro
insights applied at
pre-planning stage
View across the campaign
Step 1 to get there:
prove ROI!
Neuro methods have
the huge potential to
cover the full cycle
Which creative to launch?
How to make it better?
Which audience to buy?
How much to invest?
What was the impact?
How does it benchmark?
9. The measured social statistics
have long tail distributions.
We prefer classification to try
to predict creative excellence
instead of regression on the
raw numbers.
Although we are able to measure
different activity (view, comment, share
etc.) in reality these metrics are highly
correlated. Somewhat weaker but
positive correlations can be observed
across different data sources.
Understanding social data
10. More than 1000 features from:
• 12 measures: 6 basic emotions,
Neutral, Engagement,
Valence, Attention, Approach
and Heartrate
• Timeline features from
dynamics of emotion curves
• Event features to capture
individual behavior (e.g. % of
people with more than 3
second long smile)
Preparing the data for analysis
Volkswagen Force
70
60
50
40
30
20
10
0
1.0
0.8
0.6
0.4
0.2
0
-0.2
-0.4
Sessions
Time (0.1 sec)
0 50 100 150 200 250 300
11. Several modeling techniques
• Nearest Neighbors
• Logistic Regression
• Support Vector Machine
• Random Forest
• Gradient Boosted
Regression Trees
Happy
42%
Surprise
21%
Disgust
14%
Neutral
12%
Engagement
9%
Sad
2%
Emotion Importance
Happy Surprise Disgust Neutral Engagement Sad
Modelling approaches used for analysis
12. 0.67
0.70 0.71
0.68
0.71
0.68
0.74
0.72
0.76 0.77 0.76
0.71
0.80 0.80 0.80
0.78
0.81
0.74
FACEBOOK COMMENTS
> 5,000
FACEBOOK LIKES >
5,000
FACEBOOK SHARES >
5,000
YOUTUBE COMMENTS >
1,000
YOUTUBE LIKES > 1,000 YOUTUBE VIEWS >
1,000,000
AreaundertheROCcurve
Only self-reported features Top 12 emotion features Our best result
Performance of different approaches
13. Short Description Impact*
1 Percentage of people with smile 0.86 Happy
2 Percentage of people with long smile (>3 sec) 0.85 Surprise
3 Percentage of people with disgust 0.76 Disgust
4 Percentage of surprised people 0.73 Neutral
5 Average duration of smile events 0.69 Engagement
6 Average duration of disgust events 0.57 Sad
7 Average duration of surprise events 0.55
8 Happiness at the end 0.48
9 Engagement in the last 5 second 0.47
10 Average duration of neutral face 0.45
11 Sadness in the middle -0.19
12 Neutral in the last 5 second -0.45
*Impact is derived as the standardized group average difference between the best
ads and the rest. Positive Impact score indicates that the best ads have higher value.
Top 12 features that drive YouTube likes
15. EmotionAll® generalizes our main data science learnings into a simple 1-10 score that at any point in time is
represented relative to the whole Realeyes’ growing database of +5,000 videos.
Based mostly on our social performance work, supported by findings from analysis of 468 Cannes Lions
submissions and observation from relates academic research in the field, 4 core building blocks have emerged:
• Attract: can you grab the attention? Measured by peak surprise value early in the video.
• Retain: can you keep it? Measured by peak happiness value after the early part of the video.
• Engage: how strong engagement can you build? Measured by peak engagement anywhere in the video.
• Impact: what do you leave people with? Measured by Daniel Kahneman’s peak-to-end rule: impression left
by any experience is determined by any emotion evoked at their peak and at the end: (peak + end) / 2
EmotionAll®
16. 0M
1M
2M
3M
4M
5M
6M
7M
1 2 3 4 5 6 7 8 9 10
Views
EmotionAll® Score
Average of YouTube Views
0K
10K
20K
30K
40K
50K
60K
70K
80K
90K
1 2 3 4 5 6 7 8 9 10
Shares
EmotionAll® Score
Average of Facebook Share
Count
0k
2k
4k
6k
8k
10k
1 2 3 4 5 6 7 8 9 10
Tweets
EmotionAll® Score
Average of Twitter
Source: Realeyes analysis of 2,083 YouTube videos and 371,245 video views in March 2015
EmotionAll® in action
All outcome-linked dataset
23. 20%
25%
30%
35%
40%
45%
50%
0:00:00 0:00:10 0:00:20 0:00:30
Brand Shown Engagement Norm - Avg Engagement US 0-60s
186,826 views
M&T Bank - Chris Dambach's Story
Missing the EmotionAll® buttons
8 shares
Attract Retain Engage Impact
24. 0
2
4
6
8
10
12
14
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Days
MultipliedPerformance
Socialactionsper1000views
12x Social Actions For Heineken
Shifting media spend behind stronger
scoring videos yielded more social actions
than non-optimised distribution.
12x Social actions
(per 1000 views)
Combining Testing with Targeting
25. MeasurementTargetingTesting
Which video to launch?
How to make it better?
FinishStart
Which audience to buy?
How much to invest?
What was the impact?
How does it benchmark?
Benefits Across the Campaign
How to cover
this part?
99% of current neuro
insights applied at
pre-planning stage
29. “People will forget what you said,
people will forget what you did
but people will never forget how
you made them feel.”
Maya Angelou – Poet, author and activist
The problem we solve is that current ad measurement is not scalable and not valuable.
“Next generation survey” - replacing questionnaires and interviews is simple to understand and works