EmoBERTv2 Model

This Model Card is a work in progress and will be completed in the future (dataset upload pending, etc)

Model Description

EmoBERTv2 is a emotion text classification model trained on a large dataset of english social media posts. The model is fine-tuned from "prajjwal1-bert-tiny" EmoBERTv2 can be used for either further fine-tuning, or for usage in real-time emotion prediction applications

Datasets

This model was trained on the [Dataset Name] dataset, which is an aggregation of many datasets through relabling and data subsetting. The dataset has 9 labels: joy, sad, love, anger, disgust, surprise, neutral, fear, and worry

Training Procedure

EmoBERTv2 was fine-tuned from [Base Model Name] with specific hyperparameters [List Hyperparameters]. Training involved [X] epochs, using a learning rate of [Y].

Intended Use

This model is intended for emotion classification in [specific domains or general use]. It should be used as a tool for [Specify Applications].

Performance

EmoBERTv2 demonstrates an accuracy of 86.17% on the [Test Dataset Name]Test set. For detailed performance metrics, refer to [Link to Performance Metrics].

Bias and Fairness

While efforts have been made to reduce bias, users should be aware of potential biases in the data. It is advisable to test the model in specific contexts.

Licensing and Usage

EmoBERTv2 is released under the MIT License and can be freely used as outlined in the license.

Other Model Variations

Additional variations of EmoBERTv2 include [List Variations]. These variations offer different trade-offs in terms of size, speed, and performance.