app.py

import pandas as pd
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.neighbors import NearestNeighbors
from sentence_transformers import SentenceTransformer
import numpy as np
import streamlit as st
import os
from rapidfuzz import process
import re

def load_data(filepath):
    """Load and deduplicate the dataset."""
    data = pd.read_csv(filepath)
    data = data.drop_duplicates(subset=['book_name'], keep='first').reset_index(drop=True)
    return data

def preprocess_data(data):
    """Preprocess the dataset by cleaning text fields."""
    data['book_name'] = data['book_name'].fillna('').str.lower()
    data['summaries'] = data['summaries'].fillna('').str.lower()
    data['categories'] = data['categories'].fillna('').str.lower()
    data['combined_text'] = (
        data['book_name'] + ' ' +
        data['summaries'] * 2 + ' ' +
        data['categories']
    )
    return data

def feature_engineering(data):
    """Generate features using TF-IDF and SentenceTransformer embeddings."""
    # TF-IDF Vectorization
    vectorizer = TfidfVectorizer(stop_words='english')
    tfidf_matrix = vectorizer.fit_transform(data['combined_text'])

    # SentenceTransformer embeddings
    model = SentenceTransformer('paraphrase-MiniLM-L3-v2')
    embeddings_file = "semantic_embeddings.npy"
    if not os.path.exists(embeddings_file):
        data['semantic_embeddings'] = data['combined_text'].apply(lambda x: model.encode(x))
        np.save(embeddings_file, np.vstack(data['semantic_embeddings']))
    else:
        semantic_embeddings = np.load(embeddings_file)
        data['semantic_embeddings'] = [embedding for embedding in semantic_embeddings]

    # Fit NearestNeighbors model
    nn_model = NearestNeighbors(metric='cosine', algorithm='auto')
    nn_model.fit(np.vstack(data['semantic_embeddings']))

    return vectorizer, tfidf_matrix, model, nn_model

def calculate_similarity(input_title, data, vectorizer, tfidf_matrix, model, nn_model, n_recommendations=5, min_similarity=0.3):
    """Calculate similarities and generate recommendations."""
    try:
        # Compute TF-IDF similarity
        input_vector = vectorizer.transform([input_title.lower()])
        cosine_similarities = cosine_similarity(input_vector, tfidf_matrix).flatten()
        min_similarity = max(0.1, cosine_similarities.mean() - cosine_similarities.std())

        # Compute semantic similarity using NearestNeighbors
        input_embedding = model.encode(input_title.lower())
        distances, indices = nn_model.kneighbors([input_embedding], n_neighbors=len(data))

        # Combine similarities
        recommended_books = []
        seen_titles = set()

        for idx, dist in zip(indices[0], distances[0]):
            book_title = data['book_name'].iloc[idx]
            round_sim_value = round(1 - dist, 2)  # Convert distance to similarity
            if book_title not in seen_titles and round_sim_value >= min_similarity and str(book_title).lower() != str(input_title).lower():
                recommended_books.append((book_title, round_sim_value))
                seen_titles.add(book_title)
            if len(recommended_books) == n_recommendations:
                break

        # If no meaningful recommendations, return fallback message
        if not any(score >= min_similarity for _, score in recommended_books):
            return [("No meaningful recommendations found.", 0.0)]

        return recommended_books
    except Exception as e:
        return [(f"An error occurred: {str(e)}", 0.0)]

def validate_input(input_title, data, threshold=70):
    """Validate input title using fuzzy matching against dataset titles."""
    titles = data['book_name'].tolist()
    closest_match, score, _ = process.extractOne(input_title, titles)
    if score >= threshold:
        return closest_match
    return None

def deploy_ui(data, vectorizer, tfidf_matrix, model, nn_model):
    """Streamlit-based UI for interacting with the recommendation system."""
    st.title("Content-Based Book Recommender")
    st.write("Enter a book title to get up to 5 similar recommendations.")
    book_title = st.text_input("Enter a book title")

    if st.button("Submit"):
        # Input validation
        if len(book_title.strip()) < 3:
            st.write("Input is too short. Please enter a more meaningful book title.")
            return  

        closest_title = validate_input(book_title, data)
        
        if closest_title is None:
            st.write("No close match found. Please try a different title.")
            return 

        if closest_title:
            st.write(f"Did you mean: {closest_title}?")
            book_title = closest_title

        # Generate recommendations
        recommendations = calculate_similarity(book_title, data, vectorizer, tfidf_matrix, model, nn_model, min_similarity=0.3)
        if "error" in recommendations[0][0]:
            st.write("Error occured. Please try again later")
        
        elif recommendations is None or len(recommendations) == 0 or recommendations[0][0] == "No meaningful recommendations found.":
            st.write("No meaningful recommendations found for the input. Please try again with a different title.")
        else:
            st.write("Recommended Books:")
            idx = 1
            for book, score in recommendations:
                cleaned_book = re.sub(r"[^A-Za-z0-9]+$", "", book)
                st.write(f"{idx}: {cleaned_book}")
                idx += 1
            

if __name__ == "__main__":
    filepath = "books_summary.csv"
    data = load_data(filepath)
    data = preprocess_data(data)
    vectorizer, tfidf_matrix, model, nn_model = feature_engineering(data)
    deploy_ui(data, vectorizer, tfidf_matrix, model, nn_model)