akram-mahboub/algerian-darija-medical-dataset

# Algerian Darija Medical Triage Dataset A small, expert-labelled dataset of **859 patient symptom descriptions written in Algerian Darija**, each tagged with how urgent the case is and which medical service it belongs to. To our knowledge it is the first medical triage dataset for Algerian Darija. It was built for **Najda**, a medical triage system that reads a patient's complaint and decides (1) how soon they need to be seen and (2) which specialty should see them. --- ## What's inside Each row is one short symptom sentence with two labels: | Column | Description | Possible values | |---|---|---| | `text` | The patient's symptom description, in Algerian Darija | free text | | `urgency` | How soon the patient needs care | `Urgent`, `Consultation` | | `service` | Which medical specialty fits the complaint | `Cardiology`, `Neurology`, `Emergency Medicine`, `Internal Medicine`, `Gynecology` | **Urgency** is binary, guided by one simple question: *does this person need to be seen today, or can the visit be planned?* - `Urgent` → needs to be seen the same day. - `Consultation` → can be scheduled normally. --- ## The numbers - **859 sentences** — no empty rows, no duplicates. - **Median length:** 11 words / 51 characters (these are short, real complaints). - **Urgency split:** 445 `Urgent` / 414 `Consultation` (nearly balanced). - **Services:** roughly comparable in size, with Gynecology the smallest. How urgency lines up with each service: - **Emergency Medicine** — 100% Urgent - **Internal Medicine** — ~75% Consultation - **Cardiology** and **Gynecology** — about 60/40 - **Neurology** — almost 50/50 (neurological symptoms in Darija can be either routine or time-critical with very similar wording) --- ## Where the data comes from Two sources, mostly real text: 1. **Public Facebook health groups** for Algerian patients, where people describe symptoms in their own words — short, colloquial, code-switched, sometimes in Arabizi (Latin letters with numbers). …