File size: 2,781 Bytes
8618a3a b78fc16 8618a3a 5e2073d 8618a3a 5e2073d 8618a3a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 |
---
license: mit
tags:
- text-classification
- multitask
- toxicity
- misandry
- misogyny
- offensive-language
- english
- slovak
- xlm-roberta
---
# 🛡️ LexiGuard: Misogyny, Misandry & Toxicity Detection in English and Slovak
**LexiGuard** is a multilingual multitask model designed to detect and classify offensive language, with a focus on **misogyny**, **misandry**, and **toxicity levels** in **English**. The model also supports **Slovak**, making it suitable for multilingual analysis of social media content.
It performs **dual classification**:
1. **Category**: Misogyny, Misandry, or Neutral
2. **Toxicity level**: Low, Medium, or High
The model is based on [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) and was fine-tuned on a custom dataset primarily in **English**, with additional annotated samples in **Slovak**.
---
## 🧠 Model Overview
- **Base model**: `xlm-roberta-base`
- **Tasks**: Multitask classification (2 output heads)
- **Primary language**: English
- **Secondary language**: Slovak
- **Use case**: Detecting offensive, sexist, or toxic comments in multilingual social media
---
## 🛠️ Usage
```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification
tokenizer = AutoTokenizer.from_pretrained("Megyy/lexiguard")
model = AutoModelForSequenceClassification.from_pretrained("Megyy/lexiguard")
text = "Women are useless in politics."
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
# outputs.logits contains predictions for both tasks
```
> Note: The model has **two output heads**:
> - Head 1: Category (misogyny/misandry/neutral)
> - Head 2: Toxicity (low/medium/high)
---
## 📊 Label Definitions
**Task 1 – Category Classification**
- `0`: Neutral
- `1`: Misogyny
- `2`: Misandry
**Task 2 – Toxicity Prediction**
- `0`: Low
- `1`: Medium
- `2`: High
---
## 🧪 Training Data
- Over 5,000 manually annotated comments
- Domain: Online discussions, social media, and forums
- Language distribution:
- ~80% English
- ~20% Slovak
---
## 📁 Model Files
- `pytorch_model.bin` / `model.safetensors`: model weights
- `config.json`: model configuration
- `tokenizer.json`, `vocab.txt`, etc.: tokenizer files
- `README.md`: model card
---
## 📚 Citation
If you use this model in your work, please cite:
```
@bachelorsthesis{majercakova2025lexiguard,
title={LexiGuard: Offensive Language Detection in English and Slovak Social Media},
author={Magdalena Majercakova},
year={2025},
note={Bachelor's thesis, TUKE},
}
```
---
## 👨💻 Author
Developed by **Magdaléna Majerčáková** as part of a Bachelor's Thesis
Supervised by **Ing. Zuzana Sokolová, PhD**
Faculty of Electrical Engineering and Informatics, TUKE (2025)
---
|