SetFit Synthetic Dataset

dataset.py

# Healthcare Feedback Dataset for SetFit
# 90 examples (15 per class) - suitable for few-shot learning

# Optimized label names for ML
LABEL_MAPPING = {
    0: "access_availability",
    1: "information_provision", 
    2: "privacy_confidentiality",
    3: "continuity_care",
    4: "clinical_communication",
    5: "admin_communication"
}

# Training dataset
healthcare_dataset = [
    # access_availability (0)
    {"text": "Tried to book an appointment but earliest slot was 3 weeks away", "label": 0},
    {"text": "Phoned at 9am and got a same-day appointment - brilliant service", "label": 0},
    {"text": "Parking is a nightmare at this clinic, always full and no disabled spaces", "label": 0},
    {"text": "The wheelchair ramp has been broken for weeks with no signage", "label": 0},
    {"text": "Walk-in hours are very convenient for working professionals", "label": 0},
    {"text": "Waited 45 minutes past my scheduled appointment time", "label": 0},
    {"text": "Online booking system is intuitive and available 24/7", "label": 0},
    {"text": "They close at 5pm which is impossible for those with full-time jobs", "label": 0},
    {"text": "Receptionist always finds me an urgent slot when my condition flares up", "label": 0},
    {"text": "Phone line is constantly engaged, took 20 attempts to get through", "label": 0},
    {"text": "Telemedicine option saved me a 2-hour round trip", "label": 0},
    {"text": "Building is impossible to find with poor signage from the main road", "label": 0},
    {"text": "The triage system works well for urgent cases but routine waits are 4+ weeks", "label": 0},
    {"text": "Weekend and evening appointments make managing care so much easier", "label": 0},
    {"text": "Waiting area is overcrowded and uncomfortable for long waits", "label": 0},

    # information_provision (1)
    {"text": "Doctor took time to explain my diagnosis using diagrams I could understand", "label": 1},
    {"text": "Never received my test results, had to chase the clinic three times", "label": 1},
    {"text": "The leaflet about diabetes management was comprehensive but I needed more personalization", "label": 1},
    {"text": "Nurse explained the vaccination process step-by-step before proceeding", "label": 1},
    {"text": "Left the appointment more confused than when I arrived", "label": 1},
    {"text": "Website has excellent resources about managing chronic conditions", "label": 1},
    {"text": "Doctor used excessive medical jargon without translating for me", "label": 1},
    {"text": "Received clear written instructions for my new medication dosage", "label": 1},
    {"text": "No information provided about potential side effects of treatment", "label": 1},
    {"text": "The dietitian gave me practical nutrition advice I could actually use", "label": 1},
    {"text": "Wish they'd explain what alternative treatment options exist", "label": 1},
    {"text": "Health promotion posters in waiting area are genuinely informative", "label": 1},
    {"text": "They sent detailed prep instructions 48 hours before my procedure", "label": 1},
    {"text": "While the brochure was thorough, I still had unanswered questions", "label": 1},
    {"text": "Consultant took time to discuss lifestyle factors affecting my condition", "label": 1},

    # privacy_confidentiality (2)
    {"text": "Overheard another patient's full diagnosis details at the reception desk", "label": 2},
    {"text": "Doctors always close the door and pull the curtain for complete privacy", "label": 2},
    {"text": "Concerned about who can access my medical records through their new patient portal", "label": 2},
    {"text": "Receptionist discussed my test results within earshot of other patients", "label": 2},
    {"text": "Private consultation rooms ensure conversations remain confidential", "label": 2},
    {"text": "Not confident my data is secure with their digital check-in system", "label": 2},
    {"text": "They have a discreet area for sensitive conversations about billing", "label": 2},
    {"text": "Heard staff discussing identifiable patient details in the corridor", "label": 2},
    {"text": "The privacy screens at reception prevent others seeing my personal details", "label": 2},
    {"text": "Worried about confidentiality when leaving voicemails with symptoms", "label": 2},
    {"text": "The new open-plan pharmacy counter feels less private than the old setup", "label": 2},
    {"text": "They explicitly asked for my consent before sharing info with a specialist", "label": 2},
    {"text": "Private room made me comfortable discussing sensitive mental health issues", "label": 2},
    {"text": "Not sure if my email communications with the clinic are encrypted", "label": 2},
    {"text": "Staff are extremely discreet when handling personal information", "label": 2},

    # continuity_care (3)
    {"text": "Love that I can see Dr. Smith every time, she knows my full medical history", "label": 3},
    {"text": "Saw three different doctors in one month, had to repeat my story each time", "label": 3},
    {"text": "My GP coordinated all my specialist referrals seamlessly", "label": 3},
    {"text": "No follow-up appointment scheduled after my hospital discharge", "label": 3},
    {"text": "The care team communicates well with each other about my case", "label": 3},
    {"text": "Had to explain my rare condition from scratch to each new locum doctor", "label": 3},
    {"text": "They have a robust system for tracking ongoing chronic issues", "label": 3},
    {"text": "Dr. Jones left and no one informed me who my new named GP would be", "label": 3},
    {"text": "Excellent continuity in managing my complex multi-system condition", "label": 3},
    {"text": "The referral to physiotherapy was processed within 48 hours", "label": 3},
    {"text": "Feel like I'm starting over with every single visit", "label": 3},
    {"text": "My named nurse is consistent and knows my case inside out", "label": 3},
    {"text": "They lost my referral letter and I had to wait 4 months for a new one", "label": 3},
    {"text": "Great handover communication between hospital team and my GP", "label": 3},
    {"text": "Appreciate having the same pediatrician for my child's ongoing care", "label": 3},

    # clinical_communication (4)
    {"text": "Doctor was dismissive and interrupted me whenever I tried to speak", "label": 4},
    {"text": "Nurse was incredibly empathetic and listened without rushing me", "label": 4},
    {"text": "Felt rushed during consultation despite having multiple concerns", "label": 4},
    {"text": "The surgeon explained the procedure with patience and visual aids", "label": 4},
    {"text": "Doctor maintained eye contact and treated me as an equal partner", "label": 4},
    {"text": "Midwife was abrupt and brushed off my concerns about pain management", "label": 4},
    {"text": "Excellent bedside manner, very reassuring before my operation", "label": 4},
    {"text": "Physiotherapist used encouraging language during my rehabilitation", "label": 4},
    {"text": "Doctor seemed distracted, kept checking their phone during our session", "label": 4},
    {"text": "The anesthetist calmed my nerves with clear, compassionate explanations", "label": 4},
    {"text": "Felt like a number on a conveyor belt, not a person", "label": 4},
    {"text": "Nurse practitioner had exceptional communication skills and emotional intelligence", "label": 4},
    {"text": "Doctor was condescending when I asked questions about treatment", "label": 4},
    {"text": "The therapeutic communication was outstanding", "label": 4},
    {"text": "Consultant was approachable and made complex information accessible", "label": 4},

    # admin_communication (5)
    {"text": "Receptionist was cheerful and went above and beyond to help me", "label": 5},
    {"text": "The lady at the desk was abrupt and made me feel like a nuisance", "label": 5},
    {"text": "Booking team worked miracles to accommodate my schedule", "label": 5},
    {"text": "Administrative staff ignore patients while they chat about personal matters", "label": 5},
    {"text": "Front desk team is efficient, professional, and genuinely welcoming", "label": 5},
    {"text": "Receptionist sounded annoyed and impatient on the phone", "label": 5},
    {"text": "The admin team explains billing and insurance matters clearly", "label": 5},
    {"text": "Waited at the desk while two staff members discussed weekend plans", "label": 5},
    {"text": "Called to reschedule and received wonderfully helpful service", "label": 5},
    {"text": "Receptionist couldn't answer basic questions about forms", "label": 5},
    {"text": "Check-in process is smooth and staff maintain friendly professionalism", "label": 5},
    {"text": "The clerk was impatient when I asked for clarification on paperwork", "label": 5},
    {"text": "Administrative team handles complex queries with patience", "label": 5},
    {"text": "Desk staff were gossiping about a patient - completely unprofessional", "label": 5},
    {"text": "Appointment reminder calls are polite, clear, and not robotic", "label": 5},
]

# Convert to Huggingface Dataset
from datasets import Dataset

dataset = Dataset.from_list(healthcare_dataset)

# Optional: Create train/test split
dataset = dataset.train_test_split(test_size=0.2, seed=42)

# SetFit training example
"""
from setfit import SetFitModel, SetFitTrainer
from sentence_transformers import SentenceTransformer

# Initialize model
model = SetFitModel.from_pretrained(
    "sentence-transformers/paraphrase-mpnet-base-v2",
    labels=list(LABEL_MAPPING.values())
)

# Create trainer
trainer = SetFitTrainer(
    model=model,
    train_dataset=dataset["train"],
    eval_dataset=dataset["test"],
    metric="accuracy",
    batch_size=16,
    num_iterations=20,  # Number of text pairs to generate
    num_epochs=1,  # Number of epochs to train the classifier
)

# Train and evaluate
trainer.train()
metrics = trainer.evaluate()
"""