Measuring happiness Measuring emotional content Santa Fe - PowerPoint PPT Presentation

Happiness Some motivation Measuring happiness Measuring emotional content Santa Fe Institute, June 10, 2009 Data sets Analysis Songs Blogs Peter Dodds & Chris Danforth SOTU Future work Prediction Department of Mathematics & Statistics Center for Complex Systems References Vermont Advanced Computing Center University of Vermont Frame 1/55

Happiness Outline Some motivation Some motivation Measuring emotional content Measuring emotional content Data sets Analysis Data sets Songs Blogs SOTU Future work Analysis Prediction Songs References Blogs SOTU Future work Prediction References Frame 2/55

Happiness Happiness: Some motivation Measuring emotional content Data sets Analysis Songs ◮ Greek philosophers held Blogs SOTU Eudaimonia as highest good. [7] Future work ◮ ≃ flourishing, well-being, Prediction pleasure, ... References ◮ Socrates, Plato, Aristotle, Epicurus, ... http://wikipedia.org Frame 3/55

Happiness Happiness: Some motivation Measuring emotional content Data sets Analysis Songs Bentham’s hedonistic calculus: Blogs SOTU “[t]he greatest happiness of the greatest Future work number is the foundation of morals and Prediction legislation” [14] References Priestly, John Stuart Mill, ... http://wikipedia.org Frame 4/55

Happiness United States’ Declaration of Independence: Some motivation Measuring emotional content Data sets Analysis “We hold these truths to be sacred & Songs Blogs undeniable; that all men are created SOTU equal & independent, that from that Future work equal creation they derive rights Prediction References inherent & inalienable, among which are the preservation of life, & liberty, & the pursuit of happiness;” http://wikipedia.org Frame 5/55

Happiness Happiness: Some motivation Measuring emotional content Data sets Analysis Songs Blogs SOTU Even the odd modern economist Future work likes happiness: Prediction “Happiness” by Richard Layard [9] References ( ⊞ ) http://www.amazon.com Frame 6/55

Happiness What makes us happy? Some motivation Layard’s summary: Measuring emotional content Dominant factors: Data sets ◮ Family Analysis Songs relationships Blogs ◮ Health SOTU ◮ Financial situation Future work ◮ Personal Values ◮ Work Prediction ◮ Personal Freedom References ◮ Community and Friends Frame 7/55

Happiness What makes us happy? Some motivation Layard’s summary: Measuring emotional content Dominant factors: Data sets ◮ Family Analysis Songs relationships Blogs ◮ Health SOTU ◮ Financial situation Future work ◮ Personal Values ◮ Work Prediction ◮ Personal Freedom References ◮ Community and Friends Unimportant factors: ◮ Age ◮ Inherent ◮ Gender intelligence ◮ Looks ◮ Education Frame 7/55

Happiness Some motivation Measuring Desiring happiness—not just for boffins: emotional content Data sets ◮ Average people routinely report being happy is what Analysis they want most in life [9, 10] Songs Blogs SOTU Future work Prediction References Frame 8/55

Happiness Some motivation Measuring Desiring happiness—not just for boffins: emotional content Data sets ◮ Average people routinely report being happy is what Analysis they want most in life [9, 10] Songs Blogs SOTU Future work National indices of well-being: Prediction References ◮ Bhutan ◮ France ◮ Australia Frame 8/55

Happiness Emotional content Some motivation So how does one measure Measuring emotional content 1. happiness? Data sets 2. levels of other emotions? Analysis Songs Blogs SOTU Future work Prediction References Frame 9/55

Happiness Emotional content Some motivation So how does one measure Measuring emotional content 1. happiness? Data sets 2. levels of other emotions? Analysis Songs Blogs SOTU Just ask people how happy they are. Future work Prediction References Frame 9/55

Happiness Emotional content Some motivation So how does one measure Measuring emotional content 1. happiness? Data sets 2. levels of other emotions? Analysis Songs Blogs SOTU Just ask people how happy they are. Future work ◮ Experience sampling [2, 4, 3] (Csikszentmihalyi et al.) Prediction References ◮ Day reconstruction [8] (Kahneman et al.) Frame 9/55

Happiness Emotional content Some motivation So how does one measure Measuring emotional content 1. happiness? Data sets 2. levels of other emotions? Analysis Songs Blogs SOTU Just ask people how happy they are. Future work ◮ Experience sampling [2, 4, 3] (Csikszentmihalyi et al.) Prediction References ◮ Day reconstruction [8] (Kahneman et al.) But self-reporting has drawbacks... ◮ relies on memory and self-perception ◮ induces misreporting [11] ◮ costly Frame 9/55

Happiness Measuring Emotional Content We’d like to build an hedonometer: Some motivation Measuring ◮ An instrument to ‘remotely-sense’ emotional states emotional content Data sets and levels, in real time or post hoc. Analysis Songs Blogs SOTU Future work Prediction References Frame 10/55

Happiness Measuring Emotional Content We’d like to build an hedonometer: Some motivation Measuring ◮ An instrument to ‘remotely-sense’ emotional states emotional content Data sets and levels, in real time or post hoc. Analysis Songs Blogs Ideally: SOTU Future work ◮ Transparent ◮ Non-reactive Prediction ◮ Fast References ◮ Complementary to ◮ Based on written self-reported measures expression ◮ Improvable ◮ Uses human evaluation Frame 10/55

Happiness Measuring Emotional Content We’d like to build an hedonometer: Some motivation Measuring ◮ An instrument to ‘remotely-sense’ emotional states emotional content Data sets and levels, in real time or post hoc. Analysis Songs Blogs Ideally: SOTU Future work ◮ Transparent ◮ Non-reactive Prediction ◮ Fast References ◮ Complementary to ◮ Based on written self-reported measures expression ◮ Improvable ◮ Uses human evaluation Some possibilities: ◮ Natural language processing (e.g., OpinionFinder) ◮ Declared mood levels in blogs (e.g., Livejournal) [12] Frame 10/55

Happiness Measuring Emotional Content Some motivation Measuring emotional content ◮ Idea: Gauge emotional content of an entity through Data sets human assessment via semantic differentials. Analysis Songs Blogs SOTU Future work Prediction References Frame 11/55

Happiness Measuring Emotional Content Some motivation Measuring emotional content ◮ Idea: Gauge emotional content of an entity through Data sets human assessment via semantic differentials. Analysis ◮ Examples: Songs Blogs ◮ hate ↔ love SOTU ◮ rough ↔ smooth Future work ◮ up ↔ down Prediction References Frame 11/55

Happiness Measuring Emotional Content Some motivation Measuring emotional content ◮ Idea: Gauge emotional content of an entity through Data sets human assessment via semantic differentials. Analysis ◮ Examples: Songs Blogs ◮ hate ↔ love SOTU ◮ rough ↔ smooth Future work ◮ up ↔ down Prediction ◮ Osgood et al. (1957) [13] identified References a basis of 3 semantic differentials: ◮ Valence: bad ↔ good ◮ Dominance: weak ↔ strong ◮ Arousal: passive ↔ active (also often: Evaluation, Potency, and Activity) Frame 11/55

Happiness ANEW study Some motivation Measuring ◮ ANEW = “Affective Norms for English Words” emotional content Data sets Analysis Songs Blogs SOTU Future work Prediction References Frame 12/55

Happiness ANEW study Some motivation Measuring ◮ ANEW = “Affective Norms for English Words” emotional content Data sets Analysis ◮ Study: participants shown lists of isolated words Songs Blogs ◮ Asked to grade each word’s valence, arousal, and SOTU Future work dominance level Prediction ◮ Integer scale of 1–9 References Frame 12/55

Happiness ANEW study Some motivation Measuring ◮ ANEW = “Affective Norms for English Words” emotional content Data sets Analysis ◮ Study: participants shown lists of isolated words Songs Blogs ◮ Asked to grade each word’s valence, arousal, and SOTU Future work dominance level Prediction ◮ Integer scale of 1–9 References ◮ N = 1034 words—previously identified as bearing emotional weight ◮ Participants = College students (*cough*) ◮ Results published by Bradley and Lang (1999) [1] Frame 12/55

Happiness ANEW study—three 1–9 scales: valence: Some motivation Measuring emotional content Data sets Analysis Songs Blogs SOTU Future work Prediction References Frame 13/55

Happiness ANEW study—three 1–9 scales: valence: Some motivation Measuring emotional content Data sets Analysis Songs Blogs SOTU arousal: Future work Prediction References dominance: Frame 13/55

Happiness ANEW study: Some motivation Measuring emotional content Data sets ◮ Valence scale presented to participants as a Analysis ‘happy-unhappy scale.’ Songs Blogs SOTU Future work Prediction References Frame 14/55