Beyond Positive Emotion: Deconstructing Happy Moments Based on Writing Prompts

Kokil Jaidka, Niyati Chhaya, Saran Mumick, Matthew Killingsworth, Alon Y. Halevy, Lyle H. UngarICWSM 2020

Abstract

Happiness is often oversimplified in computational linguistics as mere positive sentiment. This research deconstructs happiness into the psychological dimensions of agency (personal mastery) and sociality (social connection). Using the CL-Aff HappyDB dataset, the authors demonstrate that these constructs can be automatically detected in text and provide a more accurate reflection of long-term life satisfaction than traditional sentiment analysis.

Introduction: Beyond Sentiment

In the field of Natural Language Processing (NLP), "happiness" is frequently measured through keyword searches or sentiment scores. However, these methods often capture fleeting moods or "social media posturing" rather than stable psychological well-being. This paper addresses this gap by identifying two pillars of happiness derived from psychological theory: Agency and Sociality.

The Pillars of Happiness

The researchers propose that true well-being is rooted in how individuals perceive their control over their lives and their connections to others.

Agency: The Sense of Mastery

Agency refers to the sense of personal control and achievement. It is characterized by moments where an individual accomplishes a goal or initiates a new phase of life.

Word cloud illustrating Agency-related terms like "job," "new," and "started."

Standard psycholinguistic tools like LIWC often lack a direct conceptual equivalent for Agency, making its automatic detection a significant contribution of this work. Linguistic analysis shows that Agency is strongly associated with verbs of action and nouns related to career and personal milestones.

Sociality: The Sense of Belonging

Sociality encompasses the sense of belonging and interaction with others. The study confirmed that social experiences—such as spending time with family or friends—are reported to have a longer duration of happiness compared to solitary activities like eating or watching movies.

Word cloud illustrating Sociality-related terms like "family," "friends," and "husband."

Methodology

The study utilized a multi-stage approach combining crowdsourcing and machine learning:

  1. Data Collection: Over 17,000 "happy moments" were collected via Amazon Mechanical Turk and annotated for Agency and Sociality.
  2. Classification: Supervised models, including Gradient Boosted Classifiers, were trained using GloVe embeddings and LIWC features. The models achieved high performance, with an AUC of 0.85 for Agency and 0.94 for Sociality.
  3. Validation: The models were applied to the myPersonality Facebook dataset to correlate linguistic scores with users' self-reported "Satisfaction With Life Scale" scores.

Key Findings

The research yielded several critical insights:

  • Predictive Validity: While standard positive emotion scores showed a negative correlation with actual life satisfaction in the Facebook dataset, the Agency model showed a positive correlation. This suggests that "feeling good" in a sentiment sense does not always equate to being satisfied with one's life.
  • Topical Mismatch: There is a notable difference between what people post on social media (e.g., #joy on Twitter often relates to romance or careers) and what they report in private writing prompts (dominated by family and food).
  • Longevity: Social joy is more enduring than solitary joy, reinforcing the psychological importance of connection.

Conclusion

By moving beyond simple sentiment and focusing on agency and sociality, researchers can better understand the underlying drivers of human well-being. This approach has significant implications for mental health monitoring and the development of more empathetic AI systems that recognize the difference between a fleeting mood and a meaningful life.

Summary updated

BibTeX
@inproceedings{jaidka2020beyond,
  author = {Kokil Jaidka and Niyati Chhaya and Saran Mumick and Matthew Killingsworth and Alon Y. Halevy and Lyle H. Ungar},
  title = {{Beyond Positive Emotion: Deconstructing Happy Moments Based on Writing Prompts}},
  booktitle = {ICWSM},
  year = {2020},
}
Figure