Mining the Relationship between Emoji Usage Patterns and Personality

Weijian Li, Yuxiao Chen, Tianran Hu, Jiebo Luo

TL;DR

This paper investigates the link between the "Big Five" personality traits and emoji usage patterns by analyzing a massive dataset of 75 million tweets. The researchers demonstrate that personality significantly influences both the frequency and specific types of emojis users choose, and that integrating personality data into machine learning models improves the accuracy of emoji prediction tasks.

What Problem Does This Paper Address?

While emojis have become a universal "non-verbal" language in digital communication, usage patterns vary wildly between individuals. Some people pepper every sentence with icons, while others rarely use them; some prefer hearts, while others stick to classic smileys.

Previous research often focused on the context of the message (sentiment, linguistic effects) to predict emoji usage. However, this paper addresses the "who" behind the screen. The authors hypothesize that human personality—specifically the "Big Five" traits—is a hidden driver behind these patterns. By moving beyond small-scale surveys to a large-scale computational approach, they aim to provide a data-driven understanding of how our inner traits manifest in our digital hieroglyphs.

Methods

The researchers employed a fully computational pipeline to bridge the gap between psychology and data science:

  1. Personality Extraction: Instead of using subjective questionnaires, the authors used the Big Five personality model (Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism). They utilized the Receptiviti tool (based on LIWC) to calculate these traits directly from the linguistic patterns in users' tweet histories.
  2. Statistical Analysis: To understand usage frequency, they compared the top and bottom quartiles of users for each trait using two-tailed t-tests. To understand preference, they calculated Spearman’s Rank Correlation Coefficients between personality scores and the usage frequency of the 67 most common emojis.
  3. Emoji Prediction: They built a machine learning classifier to predict which of the 19 most frequent emojis a user would likely use. They compared a baseline model (using only text features) against models that "fused" text features with personality scores to see if the psychological data provided additional "discriminant power."

Datasets

The study utilized a massive, custom-built Twitter dataset:

  • Initial Collection: 352,245 users and 1.13 billion tweets collected between March and June 2016.
  • Filtering: To ensure data quality, the authors removed public/bot accounts (based on follower/followee ratios), non-English tweets, and users with fewer than 500 tweets. They also filtered out "extreme" emoji users (those who used emojis in >95% or <5% of their tweets) to focus on the majority.
  • Final Dataset: 71,823 users and 75 million tweets. This represents one of the largest "volumetric" datasets used to study the intersection of personality and emoji usage.

Key Insights and Ideas

The study revealed distinct "emoji signatures" for different personality types:

  • The Introvert’s Visual Language: Surprisingly, users with low extraversion scores used emojis more frequently. The authors suggest introverts may prefer implicit visual contexts over explicit text to express themselves.
  • The Disciplined Minimalist: High Conscientiousness (self-discipline) correlates with lower emoji usage. These users tend to stick to traditional, "plainer" text to maintain a sense of vigilance and emotional control.
  • The Empathetic Agreeable: Users high in Agreeableness heavily favor heart-shaped emojis and positive faces, using them to convey empathy and social compliance that words alone might struggle to capture.
  • The Exaggerated Neurotic: Those high in Neuroticism (emotional instability) prefer exaggerated facial expressions like the "Loudly Crying Face" (😭) or "Weary Face" (😩), likely reflecting their more volatile emotional states.
  • The Openness Outlier: Openness (willingness to try new things) was the only trait that showed almost no correlation with emoji frequency. The authors posit that because emojis are now so ubiquitous, using them no longer signals a "willingness to accept new things."

Key Results

  • Frequency Significance: T-tests confirmed that four out of the five personality dimensions (all except Openness) have a statistically significant impact on how often a user employs emojis.
  • Prediction Boost: Adding personality traits as features improved the emoji classifier's accuracy. The baseline text-only model achieved 15.8% accuracy, while the model fusing text with all five personality traits reached 17.2%.
  • Trait Utility: Conscientiousness provided the most useful information for the prediction task, while Openness provided the least, reinforcing the statistical findings.

Limitations and Future Work

The authors acknowledge several areas for further exploration:

  • Topic Sensitivity: A user's personality might be stable, but their emoji usage might shift depending on the topic (e.g., discussing politics vs. personal life). Future work could look at how personality influences emoji choice within specific contexts.
  • Demographic Factors: The study does not deeply account for how age, gender, or cultural background might interact with personality to influence emoji usage.
  • Recommendation Systems: The results suggest that personality-aware emoji recommendation systems (e.g., in smartphone keyboards) could significantly improve user experience by tailoring suggestions to the user's psychological profile.

Summary updated

Abstract

Emojis have been widely used in textual communications as a new way to convey nonverbal cues. An interesting observation is the various emoji usage patterns among different users. In this paper, we investigate the correlation between user personality traits and their emoji usage patterns, particularly on overall amounts and specific preferences. To achieve this goal, we build a large Twitter dataset which includes 352,245 users and over 1.13 billion tweets associated with calculated personality traits and emoji usage patterns. Our correlation and emoji prediction results provide insights into the power of diverse personalities that lead to varies emoji usage patterns as well as its potential in emoji recommendation tasks.
BibTeX
@inproceedings{li2018mining,
  author = {Weijian Li and Yuxiao Chen and Tianran Hu and Jiebo Luo},
  title = {{Mining the Relationship between Emoji Usage Patterns and Personality}},
  booktitle = {ICWSM},
  year = {2018},
  url = {https://openreview.net/forum?id=t8hzbGmyxm},
}
Figure