metadata

language: en
tags:
  - vad
  - emotion
  - bert
license: mit
model-index:
  - name: vad-bert
    results: []
datasets:
  - reallycarlaost/emobank
base_model:
  - google-bert/bert-base-uncased

vad-bert

A BERT-based model fine-tuned to predict Valence, Arousal, and Dominance (VAD) values from text.

Intended use

This model is intended for regression tasks on emotional dimensions. It outputs 3 float values corresponding to:

Valence (pleasant vs unpleasant)
Arousal (calm vs excited)
Dominance (controlled vs in control)

Example

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("RobroKools/vad-bert")
model = AutoModelForSequenceClassification.from_pretrained("RobroKools/vad-bert")

inputs = tokenizer("I'm feeling great!", return_tensors="pt")
outputs = model(**inputs)

vad = outputs.logits.detach().squeeze().tolist()
print(vad)  # [valence, arousal, dominance]