Persian WordCloud Generation with BERTopic and ParsBERT

This tool is designed to process Persian text content (e.g., Telegram messages) and generate a word cloud from the entire dataset. It uses the BERTopic model along with UMAP and the ParsBERT language model for topic modeling and extracting common n-grams.

---

Prerequisites

Before running the project, make sure Python 3.7 or higher is installed on your system.

Install dependencies

pip install -r requirements.txt

Recommended requirements.txt content:

pandas
bertopic
umap-learn
torch
transformers
wordcloud
matplotlib
nltk

You also need a Persian font such as Vazir-Bold.ttf in the project folder or the path specified in the code.

---

File Structure

.
├── main.py                    # Entry point of the program
├── generate_wordcloud.py      # Main code for wordcloud generation
├── embedder.py                # ParsBERTEmbedder class for vectorization
├── preprocess.py              # Text preprocessing functions
├── utils.py                   # n-gram and PMI weighting calculations
├── wordclouds/                # Output folder for images
└── README.md                  # This guide file

---

Expected Input

A CSV file with at least one column named:

• txtContent — containing Persian text content

Example:

txtContent
This is a test text for analysis.
Various topics are seen in messages.

---

Run the Program

python main.py /path/to/file.csv

After execution, a word cloud image will be saved at wordclouds/wordcloud_all_data.png.

---

Notes

• Uses ParsBERT for embedding.
• Common 2-gram words are extracted and weighted based on PMI and topic distribution.
• Output is in Persian and uses a suitable font.

---

Example Output

The image below is an example of the output generated by the program: