# Negative and Positive Body‐Related Emotions Derived From Voice Recordings During a Mirror Task in Anorexia and Bulimia Nervosa: A Natural Language Processing Approach Using RoBERTa

## TL;DR

This study uses AI to analyze emotions in voice recordings of women with eating disorders during a mirror task, finding more negative and fewer positive emotions compared to healthy women.

## Contribution

The study introduces a novel use of natural language processing (RoBERTa) to detect body-related emotions in eating disorders.

## Key findings

- Women with anorexia and bulimia expressed more negative emotions like anger and disgust compared to healthy women.
- Positive emotions like admiration and joy were significantly less frequent in eating disorder groups.
- Disgust was the strongest predictor of eating disorder symptom severity across all groups.

## Abstract

Body dissatisfaction has been linked to negative and positive emotions. The validity of self‐report methods to assess emotions in individuals with eating disorders is limited, prompting a shift towards methods like natural language processing to analyze speech content. Using artificial intelligence, this study aimed to identify specific body‐related emotions elicited by looking at oneself in the mirror in individuals with eating disorders.

Women with anorexia nervosa (n = 24), women with bulimia nervosa (n = 36), and healthy women (n = 72) completed a three‐minute mirror exposure task, verbally expressing their body‐related emotions while viewing themselves in a three‐winged mirror wearing underwear. N = 132 audio recordings were transcribed and analyzed using the GermanEmotions model (based on RoBERTa) to identify 28 emotions. Univariate ANOVAs and post hoc tests were conducted to identify group differences in expressed emotions. Prediction of symptom severity was analyzed across groups.

Compared to healthy women, those with anorexia and bulimia nervosa expressed anger, disappointment, disgust, embarrassment, fear, grief, nervousness, remorse, and sadness significantly more frequently (all ds = |0.53|–|1.11|), and admiration, approval, joy, love, optimism, and pride significantly less frequently (all ds = |0.64|–|1.12|). Disgust predicted higher eating disorder symptom severity across all groups (p < 0.001).

A distinct range of body‐related emotions differentiates women with anorexia and bulimia nervosa from healthy women. Elicited negative emotions, especially higher disgust, and diminished positive emotions suggest that body‐related interventions could benefit from fostering positive emotions instead of merely reducing negative emotions.

This study analyzed voice recordings collected during a standardized mirror task in women with anorexia or bulimia nervosa, and healthy women, employing artificial intelligence to extract individual negative and positive emotions.Utilizing a state‐of‐the‐art natural language processing approach, we identified specific heightened negative emotions and diminished positive emotions in women with anorexia or bulimia nervosa versus healthy women.Disgust appeared to be the most prominent emotion across all groups.

This study analyzed voice recordings collected during a standardized mirror task in women with anorexia or bulimia nervosa, and healthy women, employing artificial intelligence to extract individual negative and positive emotions.

Utilizing a state‐of‐the‐art natural language processing approach, we identified specific heightened negative emotions and diminished positive emotions in women with anorexia or bulimia nervosa versus healthy women.

Disgust appeared to be the most prominent emotion across all groups.