app.py

import os
from groq import Groq
import gradio as gr
import numpy as np
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import sqlite3
import pandas as pd
from tqdm import tqdm

# Get the Groq API key from environment variables (in Hugging Face, this is stored as a secret)
client = Groq(
    # This is the default and can be omitted
    api_key=os.environ.get("GROQ_API_KEY"),
)


con = sqlite3.connect("file::memory:?cache=shared", check_same_thread=False)
con.row_factory = sqlite3.Row
cur = con.cursor()

# create table if not exists

cur.execute("""
            CREATE TABLE IF NOT EXISTS places (
                Place_Id INTEGER PRIMARY KEY,  -- SQLite auto-increments INTEGER PRIMARY KEY automatically
                Place_Name TEXT NOT NULL,      -- SQLite uses TEXT instead of VARCHAR
                Description TEXT,
                Category TEXT,
                City TEXT,
                Price REAL,                    -- SQLite uses REAL instead of DECIMAL or FLOAT
                Rating REAL,
                Embedding TEXT
            );
            """)


data = pd.read_csv('tourism_place.csv')


# check if the table is empty
cur.execute("SELECT * FROM places")

if cur.fetchone() is None:
    # Store the places in the database
    for i in tqdm(range(len(data))):
        cur.execute("""
            INSERT INTO places (Place_Name, Description, Category, City, Price, Rating) 
            VALUES (?, ?, ?, ?, ?, ?)
            """, (data['Place_Name'][i], data['Description'][i], data['Category'][i], data['City'][i], float(data['Price'][i]), float(data['Rating'][i]))
    )

    # Commit the changes to the database
    con.commit()

# Compute and store embeddings
def compute_and_store_embeddings():
    model = SentenceTransformer('paraphrase-MiniLM-L6-v2')  
    
    # Select all places from the database
    cur.execute("SELECT Place_Id, Place_Name, Category, Description, City FROM places")
    places = cur.fetchall()
    
    for place in places:
        # Combine PlaceName, Category, Description, and City into one string
        text = f"{place[1]} {place[2]} {place[3]} {place[4]}"
        
        # Generate embedding for the combined text
        embedding = model.encode(text)
        
        # Convert embedding to a string format to store in the database
        embedding_str = ','.join([str(x) for x in embedding])
        
        # Update the place in the database with the embedding
        cur.execute(
            "UPDATE places SET Embedding = ? WHERE Place_Id = ?", 
            (embedding_str, place[0])
        )
    
    # Commit the changes to the database
    con.commit()
# Run the function to compute and store embeddings
compute_and_store_embeddings()


# Load Hugging Face model for generating embeddings
model = SentenceTransformer('paraphrase-MiniLM-L6-v2')

# Normalize user query using Groq VM
def normalize_query(user_query):
    try:
        response = client.chat.completions.create(
            model="llama-3.1-70b-versatile",
            temperature=0.5,
            messages=[{
                "role": "user",
                "content": f"""
                    Please analyze the query: \"{user_query}\", extract Place name, Category, Description, and City. 
                    Return the response as: "Place name, Category, Description, City".
                """
            }]
        )
        normalized_user_query = response.choices[0].message.content.split('\n')[-1].strip()
        print(f"Normalized Query: {normalized_user_query}")

        return normalized_user_query

    except Exception as e:
        print(f"Error normalizing query: {e}")
        return ""

# Generate user embedding using Hugging Face model
def get_user_embedding(query):
    try:
        return model.encode(query)
    except Exception as e:
        print(f"Error generating embedding: {e}")
        return np.zeros()  

# Find similar places based on cosine similarity
def get_similar_places(user_embedding):
    similarities = []
    # Select all places from the database
    res = cur.execute("SELECT * FROM places").fetchall()
    
    for place in res:
        embedding_str = place['Embedding']  # Assuming embeddings are stored as comma-separated strings in the database
        embedding = np.array([float(x) for x in embedding_str.split(',')])  # Convert the string back to a numpy array
        
        # Compute cosine similarity
        similarity = cosine_similarity([user_embedding], [embedding])[0][0]
        similarities.append((place, similarity))
    
    # Sort results based on similarity and then by rating
    ranked_results = sorted(similarities, key=lambda x: (x[1], x[0]['Rating']), reverse=True)
    
    # Return top places
    return ranked_results

# Main function to get top 5 destinations
def get_top_5_destinations(user_query):
    normalized_query = normalize_query(user_query)
    user_embedding = get_user_embedding(normalized_query)
    similar_places = get_similar_places(user_embedding)

    if not similar_places:
        return "Tidak ada tempat yang ditemukan."

    top_places = []
    for i, (place, similarity) in enumerate(similar_places[:10]):
        top_places.append({
            'name': place['Place_Name'],
            'city': place['City'],
            'category': place['Category'],
            'rating': place['Rating'],
            'description': place['Description'],
            'similarity': similarity
        })
    print(top_places)
    return top_places

# Generate response to user using Groq VM
def generate_response(user_query, top_places):
    try:
        # Prepare the destinations data in JSON format for the model to use directly
        destinations_data = ", ".join([
            f'{{"name": "{place["name"]}", "city": "{place["city"]}", "category": "{place["category"]}", "rating": {place["rating"]}, "description": "{place["description"]}"}}'
            for place in top_places
        ])

        # System prompt: Simplified and focused on returning only the recommendations
        system_prompt = """
            You are a tour guide assistant. Your task is to present the following tourism recommendations based on what user want and needs to the user in Bahasa Indonesia.
            - For each destination, include the name, city, category, rating, and a short description.
            - Do not provide any additional commentary.
            - Only and must only return 5 places that suitable what user wants and provided the data in a clear and concise format.
            """

        # Generate the response using the model
        response = client.chat.completions.create(
            model="llama-3.1-70b-versatile",
            temperature=0.2,
            messages=[
                {"role": "system", "content": system_prompt},  # System prompt defines behavior
                {"role": "user", "content": f"Berikut adalah rekomendasi berdasarkan data: {destinations_data}"}
            ]
        )

        # Return the response content generated by the model
        return response.choices[0].message.content
    except Exception as e:
        print(f"Error generating response: {e}")
        return "Maaf, terjadi kesalahan dalam menghasilkan rekomendasi."

# Gradio Interface - User Input and Output
def chatbot(user_query):
    # Step 1: Get the top 5 destinations
    top_places = get_top_5_destinations(user_query)

    if isinstance(top_places, str):  # Error case, e.g. "No places found"
        return top_places

    # only the first 5 element of top_places
    response = generate_response(user_query, top_places)

    return response

# Define Gradio Interface
iface = gr.Interface(
    fn=chatbot,
    inputs="text",  
    outputs="text",  
    title="Tourism Recommendation Chatbot",
    description="Masukkan pertanyaan wisata Anda dan dapatkan rekomendasi tempat terbaik!"
)

# Launch the Gradio App
if __name__ == "__main__":
    iface.launch(share=True)