app.py

import gradio as gr
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import spacy
import zipfile

# Estrai il file ZIP
zip_path = os.path.join(os.getcwd(), "en_core_web_sm.zip")
extract_dir = os.path.join(os.getcwd(), "en_core_web_sm")

if not os.path.exists(extract_dir):
    with zipfile.ZipFile(zip_path, 'r') as zip_ref:
        zip_ref.extractall(extract_dir)

# Carica il modello
nlp = spacy.load(extract_dir)

# Carica il modello SentenceTransformer
model = SentenceTransformer('sentence-transformers/all-mpnet-base-v2', device='cpu')

# Preprocessamento manuale (carica il manuale da un file o base di dati)
with open('testo.txt', 'r', encoding='utf-8') as file:
    text = file.read()

# Tokenizza il testo in frasi usando SpaCy
doc = nlp(text)
sentences = [sent.text for sent in doc.sents]  # Estrarre frasi dal testo

# Crea gli embedding per il manuale
embeddings = model.encode(sentences, batch_size=8, show_progress_bar=True)

# Funzione per ottenere le frasi più rilevanti
def find_relevant_sentences(query):
    query_embedding = model.encode([query])
    similarities = cosine_similarity(query_embedding, embeddings).flatten()

    # Filtra i risultati in base alla similitudine
    threshold = 0.5
    filtered_results = [(idx, sim) for idx, sim in enumerate(similarities) if sim >= threshold]

    # Ordina i risultati per similitudine
    filtered_results.sort(key=lambda x: x[1], reverse=True)

    # Ottieni le frasi più rilevanti
    top_n = 4
    relevant_sentences = [sentences[idx] for idx, _ in filtered_results[:top_n]]
    
    return relevant_sentences

# Interfaccia Gradio
iface = gr.Interface(
    fn=find_relevant_sentences, 
    inputs=gr.Textbox(label="Insert your query"), 
    outputs=gr.Textbox(label="Relevant sentences"),
    title="Manual Querying System",
    description="Enter a question about the machine, and this tool will find the most relevant sentences from the manual."
)

# Avvia l'app Gradio
iface.launch()