Medical dataset classification for Kurdish short text over social media

Area of Science:

• Natural Language Processing
• Medical Informatics
• Social Media Analysis

Background:

• Social media platforms like Facebook generate vast amounts of user-generated text data.
• Analyzing this data can provide insights into public health discussions and medical information dissemination.
• A dedicated dataset for Kurdish medical text is needed for specialized research.

Purpose of the Study:

• To create a labeled dataset of Kurdish comments from Facebook for medical text classification.
• To establish a benchmark for analyzing medical-related discussions in the Kurdish language.
• To facilitate the development of NLP tools for the Kurdish medical domain.

Main Methods:

• Collected 6756 comments from various Facebook pages (Medical, News, Economy, Education, Sport).
• Applied a six-step preprocessing technique to clean and refine the raw comment data.
• Labeled comments into two classes: positive (medical) and negative (non-medical) for text classification.

Main Results:

• The Medical Kurdish Dataset (MKD) comprises 6756 preprocessed comments.
• The dataset is balanced with 55% negative class (non-medical) and 45% positive class (medical).
• The labeled data is suitable for supervised machine learning models in text classification.

Conclusions:

• The Medical Kurdish Dataset (MKD) is a valuable resource for NLP research in Kurdish medical contexts.
• This dataset can aid in developing systems to identify and categorize medical information from social media.
• Further research can utilize MKD for training and evaluating medical text classification models.