Suicide Emotion Models

This is Jupyter files for training the 2 DNN models.

Both models can be accessed Here

below are the details of both models and results

First model is used to predict Suicide labels

The labels are: Anxiety, Suicide Watch, Bipolar, Depression, Off my chest

Model F1 score (Macro AVG) was 0.66, and Cross validation with 5 folds yielded Accuracy of 0.684 .

Second model is used to predict Polarity and emotions

The labels include: pos, neg, anger, fear, hopefullness, hopelessness, joy, sadness, disgust

Model F1 score (Macro AVG) was 0.56, and Cross validation with 5 folds yielded Accuracy of 0.765 .

*Calmness column was dropped during training due to low support ( rows )

Keywords Extraction and correlation matrix analysis

Both keywords along with WordCloud and correlation matrix analysis graphs are shown below

Keywords Extracted Graph

WordCloud

Correlation Matrix