README.md

---
language: multilingual
license: mit
tags:
- text-classification
- spam-detection
- content-moderation
- distilbert
- small-model
- synthetic-data
- tanaos
- artifex
base_model:
- tanaos/tanaos-spam-detection-v1
datasets:
- tanaos/synthetic-spam-detection-dataset-v1
library_name: transformers.js
pipeline_tag: text-classification
---



# tanaos-spam-detection-v1 (ONNX)


This is an ONNX version of [tanaos/tanaos-spam-detection-v1](https://huggingface.co/tanaos/tanaos-spam-detection-v1). It was automatically converted and uploaded using [this Hugging Face Space](https://huggingface.co/spaces/onnx-community/convert-to-onnx).


## Usage with Transformers.js


See the pipeline documentation for `text-classification`: https://huggingface.co/docs/transformers.js/api/pipelines#module_pipelines.TextClassificationPipeline


---


<p align="center">
    <img src="https://raw.githubusercontent.com/tanaos/.github/master/assets/logo.png" width="250px" alt="Tanaos – Train task specific LLMs without training data, for offline NLP and Text Classification">
</p>

# tanaos-spam-detection-v1: A small but performant base spam detection model

This model was created by Tanaos with the [Artifex Python library](https://github.com/tanaos/artifex).

This is a **multilingual spam detection model** (it supports 15+ languages) based on [distilbert-base-multilingual-cased](https://huggingface.co/distilbert/distilbert-base-multilingual-cased) and fine-tuned on [a synthetic dataset](https://huggingface.co/datasets/tanaos/synthetic-spam-detection-dataset-v1) to classify text as ***spam*** or ***not_spam***. It is intended to be used as a first-layer spam filter for email systems, messaging applications or any other text-based communication platform.

The following categories are considered spam:

1. Unsolicited commercial advertisement or non-commercial proselytizing.
2. Fraudulent schemes. including get-rich-quick and pyramid schemes.
3. Phishing attempts. unrealistic offers or announcements.
4. Content with deceptive or misleading information.
5. Malware or harmful links.
6. Adult content or explicit material.
7. Excessive use of capitalization or punctuation to grab attention.

---

## How to Use

### Via the Artifex library (`pip install artifex`)

```python
from artifex import Artifex

spam_detection = Artifex().spam_detection

print(spam_detection("You won an IPhone 16! Click here to claim your prize."))

# >>> [{'label': 'spam', 'score': 0.9945}]
```

### Via the Transformers library

```python
from transformers import pipeline

clf = pipeline("text-classification", model="tanaos/tanaos-spam-detection-v1")

print(clf("You won an IPhone 16! Click here to claim your prize."))

# >>> [{'label': 'spam', 'score': 0.9945}]
```

## How to fine-tune (without training data)

Use the [Artifex library](https://github.com/tanaos/artifex) to fine-tune the model to any language other than English or to custom definitions of spam by generating synthetic training data on-the-fly. Install Artifex with 

```bash
pip install artifex
```

### Fine-tune to any language

```python
from artifex import Artifex

spam_detection = Artifex().spam_detection

model_output_path = "./output_model/"

spam_detection.train(
    spam_content=[
        "Publicidad comercial no solicitada o proselitismo no comercial",
        "Esquemas fraudulentos, incluidos los de enriquecimiento rápido y los esquemas piramidales",
        "Intentos de phishing, ofertas irreales o anuncios engañosos",
        "Contenido con información engañosa o fraudulenta",
        "Malware o enlaces dañinos",
        "Contenido para adultos o material explícito",
        "Uso excesivo de mayúsculas o signos de puntuación para llamar la atención",
    ],
    language="spanish",
    output_model_path=model_output_path,
)

spam_detection.load(model_output_path)
print(spam_detection("¡Has ganado un IPhone 16! Haz clic aquí para reclamar tu premio."))

# >>> [{'label': 'spam', 'score': 0.9970}]
```

### Fine-tune to custom definitions of spam

```python
from artifex import Artifex

spam_detection = Artifex().spam_detection

model_output_path = "./output_model/"

spam_detection.train(
    spam_content=[
        "Messages that contain excessive emojis or special characters",
        "Messages that promote gambling or betting activities",
        "Messages that include unsolicited attachments or files",
        "Messages that impersonate legitimate organizations or individuals",
        "Messages that use sensationalist language to provoke reactions",
        "Messages that contain misleading links or URLs",
        "Messages that encourage illegal activities or behavior",
    ],
    output_model_path=model_output_path,
)

spam_detection.load(model_output_path)
print(spam_detection("🎉🏖️✈️ Congratulations! You've won a vacation. Click the link below to claim your prize!!!"))

# >>> [{'label': 'spam', 'score': 0.9979}]
```

## Model Description

- **Base model:** `distilbert/distilbert-base-multilingual-cased`
- **Task:** Text classification (spam detection)
- **Languages:** Multilingual (15+ languages)
- **Fine-tuning data:** A synthetic, custom dataset of spam and not spam examples.

## Training Details

This model was trained using the [Artifex Python library](https://github.com/tanaos/artifex)

```bash
pip install artifex
```

by providing the following instructions and generating 10,000 synthetic training samples:

```python
from artifex import Artifex

spam_detection = Artifex().spam_detection

spam_detection.train(
    spam_content=[
        "Unsolicited commercial advertisement or non-commercial proselytizing",
        "Fraudulent schemes, including get-rich-quick and pyramid schemes",
        "Phishing attempts, unrealistic offers or announcements",
        "Content with deceptive or misleading information",
        "Malware or harmful links",
        "Adult content or explicit material",
        "Excessive use of capitalization or punctuation to grab attention",
    ],
    num_samples=10000
)
```

## Intended Uses

This model is intended to:
- Serve as a first-layer spam filter for email systems, messaging applications, or any other text-based communication platform.
- Help reduce unwanted or harmful messages by classifying text as spam or not spam.

Not intended for:
- Use in high-stakes scenarios where misclassification could lead to significant consequences without further human review.