Table 3 Predictive power of linguistic models in the current study.

From: Head versus heart: social media reveals differential language of loneliness from depression

Features

Loneliness

Depression

r

MAE

r

MAE

Demographics

Age + gender

0.133

1.90

0.253

4.63

Language

LIWC

0.190

1.87

0.283

4.58

N-grams

0.155

1.89

0.209

4.71

Topics

0.169

1.88

0.274

4.56

BERT

0.201

1.87

0.312

4.54

Demographics + Language

Age + gender + BERT

0.204

1.87

0.321

4.52

Age + gender + LIWC + N-gram + topics + BERT

0.204

1.87

0.312

4.55

  1. r = Pearson’s correlation coefficient. N-grams = words and phrases (1–3-grams). Topics = clusters of co-occurring words extracted using the Latent Dirichlet Allocation method.
  2. MAE Mean Absolute Error, LIWC Linguistic Inquiry and Word Count, English 2015 category. BERT Bidirectional Encoder Representations from Transformers.
Back to article page