emergence of things felt

32
EMERGENCE OF THINGS FELT Harnessing the Semantic Space of Facebook Feeling Tags Chris Zimmerman Mari-Klara Stein Daniel Hardt Ravi Vatrapu

Upload: christopher-james-zimmerman

Post on 21-Jan-2018

403 views

Category:

Social Media


0 download

TRANSCRIPT

EMERGENCE OF THINGS FELT Harnessing the Semantic Space of Facebook Feeling Tags

Chris Zimmerman Mari-Klara Stein Daniel Hardt Ravi Vatrapu

EMERGENCE OF THINGS FELT

DATASET

143 Different Feelings Collected

(all public posts from April 2013-November 2014)

11,908,715 Total posts

10,290,216 Discursive (mentions)

1,618,499 Feelings-Tagged posts

55.85% feelings tagged individually (90,3919) compared with 89.70% in discursive mentions.

HARNESSING THE SEMANTIC SPACE1. To understand feelings that users choose to explicitly tag and publicly share.

2. To map the semantic space of ‘Facebook feelings’.

3. To explore how (if at all) do the user-categorized ‘Facebook feelings’ differ, on the valence and arousal dimensions, from previously theorized mappings of feelings (Russell, 1983; Scherer 2005)

4. To inform organizational practices related to social media analytics (Holsapple et al. 2014), particularly sentiment analysis (cf. Stieglitz and Dang-Xuan 2013).

5. To build better analytics tools that are able to process data on a more granular level and reveal more about user sentiment than just its polarity in terms of positivity or negativity.

BACKGROUNDFEELINGS AND EMOTIONS

1. An online experience (e.g., ‘digital emotion’) is not inferior to or less valid than an offline experience (Ellis and Tucker 2015).

2. Facebook is “allowing people to produce new and innovative emotional solutions” (Ellis and Tucker, 2015, p. 178)

3. Most studies of discrete feelings have to date relied on ‘small data’ and experimental or qualitative methods (cf. Scherer, 2005).

4. ‘Big Data’ studies and NLP techniques are popular using traditional sentiment analysis (Barnaghi et al. 2015).

• Emotion : “an episode of interrelated, synchronized changes in the states of all or most of the five organismic subsystems (cognitive, neurophysiological, motivational, motor expression and subjective feeling) in response to the evaluation of an external or internal stimulus event as relevant to major concerns of the organism” (Scherer 2005: 697).

DIMENSIONALAPPROACH

The Dimensional Approach:

(Wilhelm Wundt, 1905)

o Valence (horizontal axis)

o Arousal (vertical axis)

o Tension – often excluded

MEASURINGFEELINGS

Measurable Feeling:

An experience “that is an integral blend of hedonic (pleasure–displeasure) and arousal (sleepy–activated) values” (Russell 2003, p. 147).

Any feeling, thus, can be described as a point in the valence-arousal space (ibid.; Scherer 2005).

Russel & Scherrer’s 2D classifications are combined (right).

THE FACEBOOK DOMAIN

Feelings and Social Media•Many familiar concepts on social media are not always the same as outside of social media. • Social media facilitate AND influence the generation of feelings - for example through emotional contagion. (Kramer et al. 2014).

Platform Selection – Advantages and Limitations

(-) Public Data Only / Spam

(+) Adoption / Inhabitance

(+) Personal Nature of Network

(+) Contextual Disambiguity

Mention Post

Tagged Post

DATA COLLECTION

DATA MINING: UNDERSTANDING FACEBOOK FEELINGS

VOLUMETagged Feelings

vs discursive mentions (thin lines)

PROPORTIONAL SIGNIFICANCE

Tagged Feelings

Relative to how much they are talked about (discursively)

HAPPY

Hourly mentions over time.

FEELINGS DISCUSSED ON WEEKDAYSSunday

Monday

Tuesday

Wednesday

Thursday

Friday

Saturday

0K

1K

2K

Cha

lleng

ed

0K

10K

20K

Fres

h

0

500

1000

1500

Dru

nk

Sunday

Monday

Tuesday

Wednesday

Thursday

Friday

Saturday

0K

5K

10K

Con

fiden

t

0K

20K

40K

60KE

xcite

d

0K

1K

2K

Fed

Up

Sunday

Monday

Tuesday

Wednesday

Thursday

Friday

Saturday

0K

2K

4K

Bea

utifu

l

0

100

200

300

400

Bus

y

0K

1K

2K

3K

Hom

esic

k

Sunday

Monday

Tuesday

Wednesday

Thursday

Friday

Saturday

0

500

1000

Gru

mpy

0K

5K

10K

Frus

trate

d

0K

5K

10K

15K

Hop

eles

s

Weekday Mentions from Sunday to Saturday

FACEBOOK TAG VARIATIONS

Social Dimensions - four possibilities:

- feeling happy

- feeling happy with Ravi

- feeling happy with Ravi and Dan

- feeling happy with Mari and X others

Location Dimensions – two possibilities:

- feeling happy at Warpigs

- feeling happy in Copenhagen

FEELINGS WITH OTHERS

Average: 1.1 actors

Small Groups

2 people 142,871 8.83%

3 people 38,119 2.36%

Large Groups

4 or more 316,795 19.57%

FEELINGS ON LOCATION77,112 AT PLACE (4.76%) 19,351 IN REGION (1.20%)

MACHINE LEARNING: CREATING EMOTION DETECTION

MACHINE LEARNING

BINARY APPROACHNatural Language Processing Steps:

1. Individual Feelings Classification (44)

2. Valence and Arousal Classifiers (Binary Approach)

3. 5-Way Emotion Detection (based on parent-child hierarchies)

CLASSIFIERSArousal and Valence

FEELINGSFOLKSONOMY

Facebook Feelings Tags, as generated by the crowd.

1. feelings of excitement are the most widely shared

2. positive-aroused feelings hold the most 'gravitational pull’ in general

3. there are few motivations to express neutrally-valencedfeelings with moderate levels of arousal

4. on the valence spectrum, the most negative feeling is that of sadness, greater than disappointment, anger or even disgust

HOW CLASSIFICATIONS MEASURE UP

JUXTAPOSITIONTop-down vs Bottom-Up

1. the two-dimensional valence-arousal space of ‘Facebook feelings’ is qualitatively different from prior research (cf. Russell, 1983; Scherer, 2005);

2. yet variance between domain theorists (ibid.) is much higher than their individual variance with our empirical classification of ‘Facebook feelings’.

3. extreme and mild feelings tend to be exaggerated on Facebook;

CONSTELLATIONS

GREAT! WHAT CAN WE USE THIS FOR?

BUILDING A FEELINGS METER

Organizational Relevance­ more informative classifications (than

traditional sentiment analysis)­ better monitoring of the large conversation

streams that revolve around brands online

Owned Content Conversation • Brand Positioning • Emotional Alignment

Earned Conversation • Conversation Monitoring (via Radar)• Detection from the Crowd (via Alerts)

Feelings Meter Demo Version: cssl.cbs.dk/software/feelingsmeter)

5-Way Emotion Detection

FIFA FACEBOOK WALL

However a significant spike in collective negativity is strongly apparent.

Arousal level of the conversation remains relatively unchanged. Surge in anger occurs after FIFA executives are arrested.

Fluctuations in joy levels seem to stabilize after the event.

LEVERAGING A WISDOM OF CROWDSThe nature of Facebook data also allows us to draw on the folksonomic wisdom of the crowds via several advantages:

• The sheer volume of posts allows us to leverage the contributions of the English speaking population who volunteer feeling tags as they see appropriate.

• Facebook feelings are collected across time and space, within a natural inhabitation online. Traditional survey studies are performed at specific times and spaces, not allowing subjects to appropriate emotions in an embedded fashion at any and every point in time in their daily lives.

• Our data-driven approach from big social data has allowed the patterns in feeling tag use to emerge from millions of posts, letting the data speak for itself, and to reveal observable differences from past assumptions.

Christopher ZimmermanComputational Social Science Lab

Copenhagen Business School – ITM

twitter @socialbeit [email protected]

EMERGENCE OF THINGS FELT Harnessing the Semantic Space of Facebook Feeling Tags

Chris Zimmerman Mari-Klara Stein Daniel Hardt Ravi Vatrapu ICIS 2015, Dallas, Texas

Feelings Meter Demo Version: cssl.cbs.dk/software/feelingsmeter