Can the emotional words people recall help predict their depression, anxiety, and stress levels?

Details

Most current affect scales and sentiment analysis of written text focus on quantifying valence or sentiment, the primary dimension of emotion. Distinguishing broader, more complex negative emotions with similar valence is important for evaluating mental health.

The researchers proposed DASentimental, a semi-supervised machine-learning model for extracting depression, anxiety, and stress from written text. They trained it to identify how N = 200 sequences of recalled emotional words correlated with the recallers’ depression, anxiety, and stress measured using the Depression Anxiety Stress Scale (DASS-21).

Using cognitive network science, they modeled each recall list both as a bag-of-words (BOW) vector and as a walk through a network representation of semantic memory, specifically free associations. This approach weighted BOW entries by their centrality (degree) in semantic memory and incorporated semantic network distances into the recalled-word representations, embedding them in a cognitive representation.

This embedding produced what the researchers described as state-of-the-art, cross-validated predictions for depression (R = 0.7), anxiety (R = 0.44), and stress (R = 0.52), equivalent to previous results that used additional human data. Powered by a multilayer perceptron neural network, DASentimental offers a way to investigate how emotional distress is organized semantically.

Semantic distances between recalled words—called walk coverage—were key to estimating depression levels but redundant for estimating anxiety and stress levels. Semantic distances from “fear” improved anxiety predictions, but were redundant when the “sad–happy” dyad was considered.

The researchers also applied DASentimental to a clinical dataset of 142 suicide notes. The predicted depression and anxiety levels, classified as high or low, corresponded to differences in valence and arousal as expected from a circumplex model of affect.

The researchers discuss key directions for future research enabled by artificial intelligence that detects stress, anxiety, and depression in texts.