license: bigscience-bloom-rail-1.0
Model Card for Foody Bert
Model Details
Model Description
Foody-bert results from the second round of fine-tuning on the text classification task. Continuation of fine-tuning of senty-bert, which is fine-tuned on yelp reviews and Stanford sentiment treebank with ternary labels (neutral, positive, negative).
- Developed by: Christopher Potts, Zhengxuan Wu, Atticus Geiger, and Douwe Kiela. 2020. DynaSent: A dynamic benchmark for sentiment analysis. Ms., Stanford University and Facebook AI Research.
- Shared by [Optional]: Hugging Face
- Model type: Language model
- Language(s) (NLP): More information needed
- License: bigscience-bloom-rail-1.0
- Related Models: More information needed
- Parent Model: More information needed
- Resources for more information: - Associated Paper
Uses
Direct Use
The primary intended use is in sentiment analysis of the texts of product and service reviews, and this is the domain in which the model has been evaluated to date.
We urge caution about using these models for sentiment prediction in other domains. For example, sentiment expression in medical contexts and professional evaluations can be different from sentiment expression in product/service reviews.
Downstream Use [Optional]
More information needed
Out-of-Scope Use
More information needed
Bias, Risks, and Limitations
Significant research has explored bias and fairness issues with language models (see, e.g., Sheng et al. (2021) and Bender et al. (2021)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups.
Recommendations
We recommend careful study of how these models behave, even when they are used in the domain on which they were trained and assessed. The models are deep learning models about which it is challenging to gain full analytic command; two examples that appear synonymous to human readers can receive very different predictions from these models, in ways that are hard to anticipate or explain, and so it is crucial to do continual qualitative and quantitative evaluation as part of any deployment.
We advise even more caution when using these models in new domains, as sentiment expression can shift in subtle (and not-so-subtle) ways across different domains, and this could lead specific phenomena to be mis-handled in ways that could have dramatic and pernicious consequences.
Training Details
Training Data
The model was trained on product/service reviews from Yelp, reviews from Amazon, reviews from IMDB (as defined by this dataset), sentences from Rotten Tomatoes (as given by the Stanford Sentiment Treebank), the Customer Reviews dataset, and on subsets of the DynaSent dataset. The dataset mainly contains restaurant review data.
For extensive details on these datasets are included in the associated Paper.
Training Procedure
Preprocessing
More information needed
Speeds, Sizes, Times
More information needed
Evaluation
Testing Data, Factors & Metrics
Testing Data
More information needed
Factors
More information needed
Metrics
More information needed
Results
More information needed
Model Examination
More information needed
Environmental Impact
Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
- Hardware Type: More information needed
- Hours used: More information needed
- Cloud Provider: More information needed
- Compute Region: More information needed
- Carbon Emitted: More information needed
Technical Specifications [optional]
Model Architecture and Objective
More information needed
Compute Infrastructure
More information needed
Hardware
More information needed
Software
More information needed
Citation
BibTeX:
More information needed
APA:
@article{potts-etal-2020-dynasent,
title={{DynaSent}: A Dynamic Benchmark for Sentiment Analysis},
author={Potts, Christopher and Wu, Zhengxuan and Geiger, Atticus and Kiela, Douwe},
journal={arXiv preprint arXiv:2012.15349},
url={https://arxiv.org/abs/2012.15349},
year={2020}}
Glossary [optional]
More information needed
More Information [optional]
More information needed
Model Card Authors [optional]
Christopher Potts, Zhengxuan Wu, Atticus Geiger, and Douwe Kiela. 2020. DynaSent: A dynamic benchmark for sentiment analysis. Ms., Stanford University and Facebook AI Research, in collabertation with the Hugging Face team
Model Card Contact
More information needed
How to Get Started with the Model
Use the code below to get started with the model.
Click to expand
from transformers import AutoTokenizer, AutoModelForSequenceClassification
tokenizer = AutoTokenizer.from_pretrained("rttl-ai/foody-bert")
model = AutoModelForSequenceClassification.from_pretrained("rttl-ai/foody-bert")