Positive Perspectives with English Text Reframing
Model description
This model is a T5-base adjusted to the sentiment transfer task, where the objective is to reverse the sentiment polarity of a text without contradicting the original meaning. Positive reframing induces a complementary positive viewpoint (e.g. glass-half-full) escaping negative patterns. Based on the article arXiv:2204.02952.
How to use
The model uses one or more sentiment strategies concatenated with a sentence and will generate a sentence with the applied sentiment output. The maximum string length is 1024 tokens. Entries must be organized in the following format:
Input:
['growth']: totally fed up with this bid now! :-( haven't even thought about my presentation yet :-(
Available sentiment strategies:
Strategy | Description |
---|---|
growth | viewing a challenging event as an opportunity for the author to specifically grow or improve himself. |
impermanence | Saying that bad things don't last forever, will get better soon, and/or that other people have had similar difficulties. |
neutralizing | Replacing a negative word with a neutral word. For example, “This was a terrible day” becomes “This was a long day”. |
optimism | Focusing on things about the situation itself, at that moment, that are good (not just predicting a better future). |
self_affirmation | Talking about what strengths the author already has, or values he admires, such as love, courage, perseverance, etc. |
thankfulness | Expressing gratitude or gratitude with keywords like appreciate, happy for it, grateful for, good thing, etc. |
Usage
from transformers import pipeline
pipe = pipeline('summarization', "dominguesm/positive-reframing-en")
text = "['growth']: totally fed up with this bid now! :-( haven't even thought about my presentation yet :-("
pipe(text, max_length=1024)
Output:
# I haven't thought about my presentation yet, but I'm going to work hard to improve #my presentation, and I'll be better soon.
- Downloads last month
- 87
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.