app.py

import gradio as gr
from huggingface_hub import InferenceClient
from transformers import pipeline
import re

# Load chatbot menggunakan InferenceClient
client = InferenceClient(model="mistralai/Mistral-7B-Instruct-v0.3")

# Load model Named Entity Recognition (NER)
ner_pipeline = pipeline("ner", model="d4data/biomedical-ner-all")

# Fungsi untuk ekstraksi entitas medis dari teks
def extract_entities(text):
    entities = ner_pipeline(text)
    
    merged_entities = []
    current_word = ""
    current_entity = None
    
    for ent in entities:
        word = ent["word"]
        entity_type = ent["entity"].split("-")[-1]  # Ambil tipe entitas tanpa B- atau I-
        
        # Gabungkan token subword dengan kata sebelumnya jika diawali "##"
        if word.startswith("#"):
            current_word += word.lstrip("#")  # Hapus semua "#" di awal kata
        else:
            if current_word and current_entity:
                merged_entities.append({"word": current_word, "entity": current_entity})  # Simpan kata sebelumnya
            
            current_word = word  # Mulai kata baru
            current_entity = entity_type  # Simpan tipe entity
    
    if current_word and current_entity:  # Tambahkan kata terakhir yang sudah digabung
        merged_entities.append({"word": current_word, "entity": current_entity})
    
    
    return merged_entities

# Fungsi untuk highlight teks dan menampilkan daftar entitas yang dikenali
def highlight_text(text, entities):
    words = text.split(' ')  # Pisahkan teks menjadi daftar kata
    entity_words = {ent["word"].lower(): ent for ent in entities}  # Buat dict untuk lookup cepat

    for i, word in enumerate(words):
        clean_word = word.strip('.,!?()[]')  # Hilangkan tanda baca di sekitar kata
        lower_word = clean_word.lower()

        if lower_word in entity_words:
            words[i] = f"<span style='background-color: #ffcc80; color: black; padding: 2px; border-radius: 4px;'>{word}</span>"

    highlighted_text = ' '.join(words)  # Gabungkan kembali teks

    # Buat daftar entitas yang dikenali
    if entities:
        entity_list = "<h4>🔍 Recognized Medical Entities:</h4><ul>"
        for ent in entities:
            entity_list += f"<li><strong>{ent['word']}</strong> ({ent['entity']})</li>"
        entity_list += "</ul>"
    else:
        entity_list = "<p><em>No medical entities detected.</em></p>"

    return highlighted_text + "<br><br>" + entity_list


# Fungsi chatbot dengan NER
def chat_with_ner(message, history):
    entities = extract_entities(message)
    recognized_entities = [ent["word"] for ent in entities]  # Ambil daftar kata yang dikenali

    if recognized_entities:
        prompt = f"This text contains medical terms: {', '.join(recognized_entities)}. Please explain briefly."
    else:
        prompt = message  
    show_to_history = f"Medical Object Recognized : {', '.join(recognized_entities)}. Here are the informations about the recognized medical object."
    response = client.text_generation(prompt, max_new_tokens=100)  # Gunakan text_generation()
    highlighted_message = highlight_text(message, entities)

    history.append((show_to_history, response))  # Tambahkan ke history

    return history, highlighted_message

# Desain chatbot
with gr.Blocks() as demo:
    gr.Markdown(
        "<h1 style='text-align: center;'>🚀 Mistral AI Chatbot</h1>"
        "<p style='text-align: center;'>💬 Chat with Mistral-7B and experience advanced AI conversations!</p>",
    )

    chatbot = gr.Chatbot(label="Mistral AI Assistant")
    message = gr.Textbox(placeholder="Type your message here...", label="Your Message")
    highlighted_output = gr.HTML(label="Highlighted Text (NER)")  # Ganti ke gr.HTML agar teks ter-highlight
    send_btn = gr.Button("Send 🚀")

    def respond(user_input, chat_history):
        return chat_with_ner(user_input, chat_history)

    send_btn.click(respond, inputs=[message, chatbot], outputs=[chatbot, highlighted_output])

# Jalankan aplikasi
demo.launch()