app.py

import streamlit as st
from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline
import pandas as pd

# Memuat model dan tokenizer IndoBERT
@st.cache_resource
def load_indobert_model():
    model_name = "indobenchmark/indobert-base-p1"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForQuestionAnswering.from_pretrained(model_name)
    return pipeline("question-answering", model=model, tokenizer=tokenizer)

qa_pipeline = load_indobert_model()

# Memuat dataset SOP
@st.cache_data
def load_sop_dataset():
    return pd.read_csv("dataset.csv")

dataset = load_sop_dataset()

# Fungsi untuk menemukan konteks terbaik
def find_best_context(question, dataset):
    best_score = 0
    best_context = None
    for _, row in dataset.iterrows():
        overlap = len(set(question.lower().split()) & set(row["text"].lower().split()))
        if overlap > best_score:
            best_score = overlap
            best_context = row["text"]
    return best_context

# Antarmuka Streamlit
st.title("Sistem Penjawab Pertanyaan SOP dengan IndoBERT")
st.markdown("Ajukan pertanyaan seputar Prosedur Operasional Standar:")

question = st.text_area("Masukkan pertanyaan Anda:", "")

if st.button("Dapatkan Jawaban"):
    if question:
        with st.spinner("Mencari konteks yang relevan..."):
            context = find_best_context(question, dataset)
            if context:
                with st.spinner("Menjawab pertanyaan Anda..."):
                    result = qa_pipeline(question=question, context=context)
                    st.success("Jawaban:")
                    st.write(result["answer"])
                    st.write("Skor Keyakinan:", result["score"])
            else:
                st.warning("Konteks yang relevan tidak ditemukan. Silakan coba pertanyaan lain.")
    else:
        st.warning("Silakan masukkan pertanyaan.")