This is a fine-tuned roberta-base model trained specifically to detect 28 distinct emotional states within the context of Steam video game reviews.

Model Details

Base Model: roberta-base
Task: Text Classification (Multi-class / 28 Labels)
Language: English
Training Dataset: bramb0/synthetic-steam-emotions

Intended Use

This model is designed to perform highly nuanced emotion detection on gaming-related text, such as Steam reviews, Reddit threads, and Discord messages. It goes beyond standard Positive/Negative sentiment by predicting the probabilities of 28 distinct GoEmotion labels.

Labels

The model predicts the following 28 emotions: admiration, amusement, anger, annoyance, approval, caring, confusion, curiosity, desire, disappointment, disapproval, disgust, embarrassment, excitement, fear, gratitude, grief, joy, love, nervousness, optimism, pride, realization, relief, remorse, sadness, surprise, neutral.

Training Data

The model was trained on a custom dataset of Steam Reviews. The reviews were split into individual clauses and tagged with one of the 28 emotion labels using an LLM to generate synthetic ground-truth data.

Usage

python

from transformers import pipeline

classifier = pipeline("text-classification", model="bramb0/roberta-steam-review-emotions", return_all_scores=True)
results = classifier("The gameplay is a bit mid but overall it's a 7/10 game, it has good story and lore!")
print(results)