Hebemo Surprise

🚀 HebEMO - Emotion Recognition Model for Modern Hebrew

HebEMO is a powerful tool designed to detect polarity and extract emotions from modern Hebrew User-Generated Content (UGC). It was trained on a unique Covid-19 related dataset that we collected and annotated. The model achieved a high performance, with a weighted average F1-score of 0.96 for polarity classification. In emotion detection, it reached an F1-score ranging from 0.78 to 0.97, except for the surprise emotion, which the model had difficulty capturing (F1 = 0.41). These results outperform the best-reported performance, even when compared to English language models.

✨ Features

• Polarity and Emotion Detection: HebEMO can accurately detect the polarity and extract emotions from modern Hebrew UGC.
• High Performance: The model achieved excellent results in both polarity classification and emotion detection.
• Unique Dataset: Trained on a unique Covid-19 related dataset collected and annotated by the authors.

📦 Installation

The installation steps can be found in the usage examples section. You need to clone the repository and install the required libraries.

!git clone https://github.com/avichaychriqui/HeBERT.git
!pip install pyplutchik==0.0.7
!pip install transformers==4.14.1

💻 Usage Examples

Basic Usage

# Clone the repository and import the necessary modules
!git clone https://github.com/avichaychriqui/HeBERT.git
from HeBERT.src.HebEMO import *
HebEMO_model = HebEMO()

# Analyze text from a file
HebEMO_model.hebemo(input_path = 'data/text_example.txt')
# return analyzed pandas.DataFrame  

# Analyze text directly and plot the results
hebEMO_df = HebEMO_model.hebemo(text='החיים יפים ומאושרים', plot=True)

Advanced Usage - Sentiment Classification

from transformers import AutoTokenizer, AutoModel, pipeline

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("avichr/heBERT_sentiment_analysis") #same as 'avichr/heBERT' tokenizer
model = AutoModel.from_pretrained("avichr/heBERT_sentiment_analysis")

# Create a sentiment analysis pipeline
sentiment_analysis = pipeline(
    "sentiment-analysis",
    model="avichr/heBERT_sentiment_analysis",
    tokenizer="avichr/heBERT_sentiment_analysis",
    return_all_scores = True
)

# Analyze text
sentiment_analysis('אני מתלבט מה לאכול לארוחת צהריים')	
sentiment_analysis('קפה זה טעים')
sentiment_analysis('אני לא אוהב את העולם')

📚 Documentation

Emotion UGC Data Description

Our UGC data consists of comments posted on news articles collected from 3 major Israeli news sites between January 2020 and August 2020. The total data size is approximately 150 MB, containing over 7 million words and 350K sentences.

Around 2000 sentences were annotated by crowd members (3 - 10 annotators per sentence) for overall sentiment (polarity) and eight emotions: anger, disgust, anticipation, fear, joy, sadness, surprise, and trust. The table below shows the percentage of sentences in which each emotion appeared.

Performance

Emotion Recognition

The above metrics are for the positive class (meaning, the emotion is reflected in the text).

Sentiment (Polarity) Analysis

⚠️ Important Note

The sentiment (polarity) analysis model is also available on AWS! For more information, visit AWS' git

🔧 Technical Details

The model was trained on a unique Covid-19 related dataset, which gives it an edge in analyzing modern Hebrew UGC. The high performance in polarity classification and emotion detection is a result of the carefully designed training process and the quality of the annotated data.

📄 License

No license information is provided in the original document.

Contact Us

• Avichay Chriqui
• Inbal yahav
• The Coller Semitic Languages AI Lab

Thank you, תודה, شكرا

Citation

If you used this model, please cite us as:

Chriqui, A., & Yahav, I. (2022). HeBERT & HebEMO: a Hebrew BERT Model and a Tool for Polarity Analysis and Emotion Recognition. INFORMS Journal on Data Science, forthcoming.

@article{chriqui2021hebert,
  title={HeBERT \& HebEMO: a Hebrew BERT Model and a Tool for Polarity Analysis and Emotion Recognition},
  author={Chriqui, Avihay and Yahav, Inbal},
  journal={INFORMS Journal on Data Science},
  year={2022}
}