import os
import pandas as pd
import numpy as np
import gradio as gr
from langchain_community.document_loaders import TextLoader
from langchain_text_splitters import CharacterTextSplitter
from langchain_chroma import Chroma
from langchain_huggingface import HuggingFaceEmbeddings
# =========================================================
# BASE DIRECTORY — works both locally and on HuggingFace
# =========================================================
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
# =========================================================
# EMBEDDING MODEL
# =========================================================
embedding_model = HuggingFaceEmbeddings(
model_name="sentence-transformers/all-MiniLM-L6-v2"
)
# =========================================================
# LOAD BOOK DATASET
# =========================================================
books = pd.read_csv(os.path.join(BASE_DIR, "books_with_emotions.csv"))
# =========================================================
# CLEAN CATEGORY COLUMN
# =========================================================
books["simple_categories"] = (
books["simple_categories"]
.fillna("")
.astype(str)
.str.strip()
)
# =========================================================
# HANDLE THUMBNAILS
# =========================================================
books["thumbnail"] = books["thumbnail"].fillna("")
books["large_thumbnail"] = np.where(
books["thumbnail"] != "",
books["thumbnail"] + "&fife=w800",
"cover-not-found.png"
)
# =========================================================
# LOAD EXISTING VECTOR DATABASE
# NOTE: ChromaDB files must be inside a "chroma_db" folder
# at the root of your HuggingFace Space repo.
# Required files inside chroma_db/:
# - chroma.sqlite3
# - data_level0.bin
# - header.bin
# - length.bin
# - link_lists.bin
# - index_metadata.pickle
# =========================================================
CHROMA_PATH = os.path.join(BASE_DIR, "chroma_db")
db_books = Chroma(
persist_directory=CHROMA_PATH,
embedding_function=embedding_model
)
# =========================================================
# RETRIEVE SEMANTIC RECOMMENDATIONS
# =========================================================
def retrieve_semantic_recommendations(
query: str,
category: str,
tone: str = "All",
initial_top_k: int = 50,
final_top_k: int = 16
) -> pd.DataFrame:
recs = db_books.similarity_search(query, k=initial_top_k)
books_list = []
for rec in recs:
try:
isbn = int(rec.page_content.strip('"').split()[0])
books_list.append(isbn)
except Exception:
continue
# Preserve vector search order and deduplicate
seen = set()
valid_isbns = []
for isbn in books_list:
if isbn in books["isbn13"].values and isbn not in seen:
valid_isbns.append(isbn)
seen.add(isbn)
if len(valid_isbns) == 0:
return pd.DataFrame()
book_recs = (
books
.set_index("isbn13")
.loc[valid_isbns]
.reset_index()
)
# Category filter
if category != "All":
book_recs = book_recs[
book_recs["simple_categories"].str.lower()
== category.strip().lower()
]
# Emotional tone filter
tone_map = {
"Happy": "joy",
"Surprising": "surprise",
"Angry": "anger",
"Suspenseful": "fear",
"Sad": "sadness",
}
if tone in tone_map:
book_recs = book_recs.sort_values(
by=tone_map[tone], ascending=False
)
return book_recs.head(final_top_k)
# =========================================================
# GENERATE HTML BOOK CARDS
# =========================================================
def recommend_books(query: str, category: str, tone: str):
if not query.strip():
return """
💡 Please enter a book description to get started.
"""
recommendations = retrieve_semantic_recommendations(
query=query,
category=category,
tone=tone
)
if recommendations.empty:
return """
😔 No books found for this combination.
"""
html = """
"""
for _, row in recommendations.iterrows():
title = str(row["title"])
authors = str(row["authors"])
description = str(row["description"])
short_description = " ".join(description.split()[:35]) + "..."
authors_split = authors.split(";")
if len(authors_split) == 2:
authors_str = f"{authors_split[0]} and {authors_split[1]}"
elif len(authors_split) > 2:
authors_str = (
f"{', '.join(authors_split[:-1])} and {authors_split[-1]}"
)
else:
authors_str = authors
google_books_link = (
f"https://books.google.com/books?vid=ISBN{row['isbn13']}"
)
amazon_link = (
f"https://www.amazon.com/s?k={title.replace(' ', '+')}"
)
html += f"""
{title}
by {authors_str}
{short_description}
"""
html += "