README.md

Animal Advocacy Prediction Models

Overview

A suite of prediction models designed to evaluate animal advocacy content through two complementary approaches:

1. Performance Prediction – Evaluates content based on real-world advocacy metrics (i.e. social media and email marketing analytics, sentiement analysis of comments and replies etc.).
2. Preference Prediction – Assesses content using human and synthetic feedback (i.e. assessing the cultural sensitivity, emotional impact and persuasiveness of animal rights messages).

👉 Browse our full suite of text prediction models on HuggingFace

Using the Models

These models are text regression models that predict various metrics for animal advocacy content. You can easily use them through the Hugging Face transformers library pipeline.

Using HuggingFace's Transformers Library

from transformers import pipeline

# Initialize the pipeline for text regression
predictor = pipeline(
    "text-classification",
    model="open-paws/perceived_trustworthiness_prediction",
    return_all_scores=True
)

# Make predictions
text = "Animals deserve to live free from exploitation and suffering."
result = predictor(text)

# Results will be a score between 0 and 1
print(result[0][0]['score'])  # Example output: 0.856

Batch Processing

For processing multiple texts efficiently:

texts = [
    "Animals deserve to live free from exploitation.",
    "Factory farming causes immense suffering.",
    "We must take action to protect animals."
]

# Process multiple texts at once
results = predictor(texts)

# Access scores
for text, result in zip(texts, results):
    score = result[0]['score']
    print(f"Text: {text}\nScore: {score}\n")

Known Limitations and Best Practices

Handling Out-of-Range Predictions

These models are regression-based and can occasionally predict scores outside the intended 0-1 range. This occurs when the model encounters content that it predicts should score more extremely than examples from its training data. To handle this gracefully, we recommend clipping the values to the valid range:

from transformers import pipeline
import numpy as np

predictor = pipeline(
    "text-classification",
    model="open-paws/perceived_trustworthiness_prediction",
    return_all_scores=True
)

def clip_score(score):
    """Clips the score to be between 0 and 1."""
    return np.clip(score, 0, 1)

text = "Animals deserve to live free from exploitation and suffering."
result = predictor(text)
score = result[0][0]['score']
clipped_score = clip_score(score)

# For batch processing
texts = [
    "Animals deserve to live free from exploitation.",
    "Factory farming causes immense suffering.",
    "We must take action to protect animals."
]

results = predictor(texts)
for text, result in zip(texts, results):
    score = clip_score(result[0]['score'])
    print(f"Text: {text}\nClipped Score: {score}\n")