Ah7med/BERTopic_ArXiv

tags:

• —bertopic libraryname: bertopic pipelinetag: text-classification ---

-BERTopicArabnews

A modular implementation of BERTopic for topic modeling, specifically trained on Arabic news articles. This implementation allows for flexible component selection at each layer of the topic modeling pipeline.

The core of this project is BERTopic, which is used to perform topic modeling on the processed text. The following steps are performed:

- Topic Modeling: BERTopic is trained on the cleaned dataset to identify topics in the articles.

- Fine-tuning with KeyBERT: We use KeyBERT-inspired representations to improve the clarity and interpretability of the topics.

- Topic Extraction: The most frequent topics are extracted, and each document is assigned a topic.

- Topic Updates: The model can be fine-tuned by updating topics with n-grams for more domain-specific phrases.

Usage

To use this model, please install BERTopic:

pip install -U bertopic

You can use the model as follows:

from bertopic import BERTopic
topic_model = BERTopic.load("Ah7med/BERTopic_ArXiv")

topic_model.get_topic_info()

Topic overview

• —Number of topics: 92
• —Number of training documents: 30721

<details> <summary>Click here for an overview of all topics.</summary>

</details>

Dataset

The dataset used in this project is saudinewsnet. It can be loaded using the following command:

from datasets import load_dataset
dataset = load_dataset("saudinewsnet")

For more information, visit the Hugging Face dataset page

Dataset Details:

Name: saudinewsnet

Description: This dataset contains news articles in Arabic from various domains, primarily focused on Saudi news.

Source: Hugging Face dataset page

Training hyperparameters

• —calculate_probabilities: True
• —language: None
• —low_memory: False
• —mintopicsize: 10
• —ngramrange: (1, 1)
• —nr_topics: None
• —seedtopiclist: None
• —topnwords: 10
• —verbose: True
• —zeroshotminsimilarity: 0.7
• —zeroshottopiclist: None

Framework versions

• —Numpy: 1.26.4
• —HDBSCAN: 0.8.40
• —UMAP: 0.5.7
• —Pandas: 2.2.2
• —Scikit-Learn: 1.2.2
• —Sentence-transformers: 3.3.1
• —Transformers: 4.47.0
• —Numba: 0.60.0
• —Plotly: 5.24.1
• —Python: 3.10.12

Visualization: Displays topic distribution and document-level information