Pearl Pu - Emotion Detection In Social Media

17
Emotion Detection in Social Media School of Computer and Communication Sciences EPFL Pearl Pu

Transcript of Pearl Pu - Emotion Detection In Social Media

Emotion Detection in Social MediaSchool of Computer and

Communication Sciences

EPFL

Pearl Pu

Why Emotions?

• They motivate us to take actions

• They regulate decision and thought

processes

• They help us understand human behaviors

http://nobaproject.com/modules/functions-of-emotions

What is emotion?

✓emotion is a reaction to events

important to our organism

✓emotion leads to changes in

multiple organismic subsystems

Plutchik, The Nature of Emotions, 2001

Brave & Nass, Agents that care: Investigating the effects of orientation of emotion

exhibited by an embodied computer agent, 2003

Emotion = coordinated changes in

organismic subsystems

5 emotion components

Cognitive (appraisal)

Neurophysiological (bodily symptoms)

Motivational or Behavioral (action tendencies)

Motor expression (facial and vocal expressions)

Subjective feeling or Affectiv (emotional experience)

rererere

rere

rerere

rerere

rere

Ekman’s 6 emotions

Ekman, An argument for basic emotions, 1992

rererere

rere

rerere

rerere

rere

Plutchik Wheel

Plutchik, The Nature of Emotions, 2001

Geneva Emotion Wheel

PositiveNegative

High

Control

Low

Control

Skin temperature

Facial expression

Gesture

Voice

Text

How do we detect emotion?

Digimind Evolution (2003)

http://www.digimind.com/

Digimind is a social media listening and monitoring tool

leveraging on the results of emotion recognition in text. It was a

B2B company based in Paris. What do we see in this particular

example? On the upper left side: it’s comparing the different

smart phones: how users discuss, perceive these brands.

You can analyze topics and track

competitors using this tool. You can monitor and analyze sub-

topics within the larger topic. Drill down into details on what's

happening, when it's happening, where, who's talking about you

(your key influencers) and how (what's the sentiment).

11

EmotionWatch - EPFL

BA C

Emotional Categories Tweet frequencyEmotion shape

explore and investigate

the full variety of elicited

emotional reactions

Our objective

13

How does it work?

Given a piece of text, detect automatically

one or several words indicative of the

respective emotions

Possible output

One emotion Distribution of emotions Several emotions

Human Computation Task

Emotion LabelHappiness, Anger, Fear, No

emotion…

Emotion StrengthLow, Medium, High

Constructed Lexicon OlympLex

• 3193 terms with attached emotion distributions

• Examples (per quadrant)

Anger, Disgust, Scorn, …

unfair, mad, ugh, annoyed,

ticked off, idiots, slap, offended,

epicfail, …

Pride, Happiness, Interest, …

bravo, champions, my girl, hero,

woohoo, sohappy, good job,

yessss, …

ouch, noooo, eek, tough to

watch, heartbroken, feel so bad,

fearful, …

Sadness, Fear, Pity, …

astounded, luv u, incredible

talent, omg, marry me, desiring,

amaze, …

Love, Surprise, Awe, …

Detecting emotion in social media helps us

understand users’perception & attitudes

Conclusion

DISCLAIMER

Any use, republication or redistribution of this content is

expressly prohibited without the prior written consent of the

Author. Permission to copy and reproduce content may be

granted by the author, at their discretion, and by request

only.

Source: presentation of Pearl Pu, EPFL at the

2015 SITA Air Transport IT Summit, Brussels.

2015 Air Transport IT Summit