app.py

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

# Percorso del file zip
zip_path = '/mnt/data/en_core_web_sm.zip'  # o il percorso che hai trovato

# Directory in cui estrarre il modello
extraction_dir = '/mnt/data/en_core_web_sm'  # o il percorso di estrazione scelto

# Estrai il file zip
with zipfile.ZipFile(zip_path, 'r') as zip_ref:
    zip_ref.extractall(extraction_dir)
    
# Carica il modello Spacy
nlp = spacy.load(extraction_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
# 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 = 0.2
    filtered_results = [(idx, sim) for idx, sim in enumerate(similarities) if sim >= threshold]

    if not filtered_results:  # Se nessun risultato supera la soglia
        return ["No relevant sentences found."]

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

    # Limita i risultati alle top_n frasi
    top_n = 4
    relevant_sentences = [sentences[idx] for idx, _ in filtered_results[:top_n]]

    # Rimuove duplicati e segmenta in frasi
    unique_sentences = list(dict.fromkeys(relevant_sentences))  # Mantiene l'ordine
    doc = nlp(" ".join(unique_sentences))
    grouped_results = [sent.text.strip() for sent in doc.sents]

    return grouped_results


examples = [
    ["irresponsible use of the machine?"],
    ["If I have a problem how can I get help? "],
    ["precautions when using the cutting machine"],
    ["How do I change the knife of the cutting machine?"]
    ["Uso irresponsable de la máquina cortadora ?"]
]

# Interfaccia Gradio
iface = gr.Interface(
    fn=find_relevant_sentences, 
    inputs=gr.Textbox(label="Insert your query"), 
    outputs=gr.Textbox(label="Relevant sentences"),
    examples=examples, 
    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()