You Are Known by How You Vlog: Personality Impressions and Nonverbal Behavior in YouTube

Joan-Isaac Biel, Oya Aran, Daniel Gática-Pérez

Abstract

An increasing interest in understanding human perception in social media has led to the study of the processes of personality self-presentation and impression formation based on user profiles and text blogs. However, despite the popularity of online video, we do not know of any attempt to study personality impressions that go beyond the use of text and still photos. In this paper, we analyze one facet of YouTube as a repository of brief behavioral slices in the form of personal conversational vlogs, which are a unique medium for self-presentation and interpersonal perception. We investigate the use of nonverbal cues as descriptors of vloggers' behavior and find significant associations between automatically extracted nonverbal cues for several personality judgments. As one notable result, audio and visual cues together can be used to predict 34% of the variance of the Extraversion trait of the Big Five model. In addition, we explore the associations between vloggers' personality scores and the level of social attention that their videos received in YouTube. Our study is conducted on a dataset of 442 YouTube vlogs and 2,210 annotations collected using Amazon's Mechanical Turk.

Introduction

While personality research in social media has historically focused on text-based blogs and static profile pictures, the explosion of online video presents a more complex medium for self-presentation. Humans naturally make "thin-slice" judgments—rapid personality assessments based on brief observations. This study explores how these impressions are formed in the "wild" of YouTube by analyzing nonverbal cues in vlogs. The authors address the gap in understanding how nonverbal behavior (how we speak and move, rather than what we say) influences how viewers perceive a vlogger's personality and how those perceptions relate to a video's popularity.

Methodology

The researchers analyzed the first minute of 442 conversational vlogs (47% male, 53% female). The methodology involved three main components:

Nonverbal Feature Extraction

  • Audio: Statistics on speaking activity (time, turns, rate) and emphasis patterns (pitch/F0, energy) were extracted using a hidden Markov model (HMM) toolbox.
  • Visual: The study introduced Weighted Motion Energy Images (wMEI). This technique collapses a video sequence into a single grayscale image where pixel intensity represents the amount of motion in that area over time.
Example of a Weighted Motion Energy Image (wMEI) used to quantify vlogger movement.

Statistical descriptors like entropy, mean, and center of mass were calculated from these images to quantify physical activity.

Crowdsourced Annotations

Amazon Mechanical Turk (MTurk) was used to collect 2,210 personality annotations (5 per video). Workers used the Ten-Item Personality Inventory (TIPI) to measure the "Big Five" traits: Extraversion (E), Agreeableness (A), Conscientiousness (C), Emotional Stability (ES), and Openness to Experience (OE).

Key Results and Insights

The study found that nonverbal cues are powerful predictors of perceived personality.

Prediction Accuracy

The combination of audio and visual cues predicted 34% of the variance for the Extraversion trait. Audio alone accounted for 24%, while visual cues alone accounted for 14%. Extraversion was the most reliably judged trait (ICC(1,k) = 0.77), while Emotional Stability was the hardest for annotators to agree upon (ICC(1,k) = 0.42).

Nonverbal Correlations

The researchers identified significant associations between specific cues and personality judgments.

Correlations between automatically extracted audio/visual cues and perceived Big Five personality traits.
  • Extraversion is the "Loudest" Trait: It showed the strongest correlations with nonverbal cues, particularly speaking time and motion energy.
  • Motion as a Personality Proxy: High levels of motion (wMEI) were positively correlated with perceived Extraversion and Openness, but negatively correlated with Conscientiousness.

Social Attention

A striking finding was the link between perceived personality and social success. Vloggers perceived as more extraverted, open to experience, and conscientious tended to receive significantly more views. Perceived Extraversion had a massive correlation with social attention (r = 0.93), followed by Openness to Experience (r = 0.78) and Conscientiousness (r = 0.62).

Discussion

This research confirms that "thin-slice" impressions of personality in vlogs are robust and can be modeled using automated nonverbal analysis. However, several areas remain for future exploration. The study measures perceived personality rather than the vloggers' actual self-reported personality. Additionally, while the correlation between personality and views is strong, the direction of causality is unclear: does the personality attract viewers, or do popular vloggers adopt specific nonverbal "personas"? Future work could also move from regression models to automated classification of vloggers into personality types.

Summary updated

Abstract

An increasing interest in understanding human perception in social media has led to the study of the processes of personality self-presentation and impression formation based on user profiles and text blogs. However, despite the popularity of online video, we do not know of any attempt to study personality impressions that go beyond the use of text and still photos. In this paper, we analyze one facet of YouTube as a repository of brief behavioral slices in the form of personal conversational vlogs, which are a unique medium for self-presentation and interpersonal perception. We investigate the use of nonverbal cues as descriptors of vloggers' behavior and find significant associations between automatically extracted nonverbal cues for several personality judgments. As one notable result, audio and visual cues together can be used to predict 34% of the variance of the Extraversion trait of the Big Five model. In addition, we explore the associations between vloggers' personality scores and the level of social attention that their videos received in YouTube. Our study is conducted on a dataset of 442 YouTube vlogs and 2,210 annotations collected using Amazon's Mechanical Turk.
BibTeX
@inproceedings{biel2021you,
  author = {Joan-Isaac Biel and Oya Aran and Daniel Gática-Pérez},
  title = {{You Are Known by How You Vlog: Personality Impressions and Nonverbal Behavior in YouTube}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2021},
  doi = {10.1609/icwsm.v5i1.14160},
}
Figure