Table 2 The performance of different models and the contribution of different social media features and data types for alcohol use risk assessment
From: Identifying substance use risk based on deep neural networks and Instagram social media data
Model | Feature/data type | AUROC |
|---|---|---|
Logistic Regression Model (Baseline) | Face features | 0.54 |
Alcohol-related captions | 0.51 | |
Alcohol-related comments | 0.50 | |
Face features and alcohol-related captions | 0.54 | |
Face features and alcohol-related comments | 0.54 | |
Alcohol-related captions and comments | 0.51 | |
All three features combined | 0.55 | |
Our Deep-Learning Model | Images only | 0.54 |
Captions only | 0.56 | |
Comments only | 0.60 | |
Images and captions | 0.56 | |
Images and comments | 0.61 | |
Captions and comments | 0.61 | |
All three data types combined | 0.65 |
