File size: 4,041 Bytes
391de07 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 |
import pandas as pd
import numpy as np
from dotenv import load_dotenv
from langchain_community.document_loaders import TextLoader
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_text_splitters import CharacterTextSplitter
from langchain_chroma import Chroma
import gradio as gr
load_dotenv()
books = pd.read_csv("books_with_emotions.csv")
books["large_thumbnail"] = books["thumbnail"] + "&fife=w800"
books["large_thumbnail"] = np.where(
books["large_thumbnail"].isna(),
"cover-not-found.jpg",
books["large_thumbnail"],
)
raw_documents = TextLoader("tagged_description.txt", encoding = 'utf-8').load()
text_splitter = CharacterTextSplitter(separator="\n", chunk_size=0, chunk_overlap=0)
documents = text_splitter.split_documents(raw_documents)
# use opensource embedding model instead of openai
embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/paraphrase-MiniLM-L3-v2")
db_books = Chroma.from_documents(
documents,
embedding=embedding_model
)
def retrieve_semantic_recommendations(
query: str,
category: str = None,
tone: str = None,
initial_top_k: int = 50,
final_top_k: int = 16,
) -> pd.DataFrame:
recs = db_books.similarity_search(query, k=initial_top_k)
books_list = [int(rec.page_content.strip('"').split()[0]) for rec in recs]
book_recs = books[books["isbn13"].isin(books_list)].head(initial_top_k)
if category != "All":
book_recs = book_recs[book_recs["simple_categories"] == category].head(final_top_k)
else:
book_recs = book_recs.head(final_top_k)
if tone == "Happy":
book_recs.sort_values(by="joy", ascending=False, inplace=True)
elif tone == "Surprising":
book_recs.sort_values(by="surprise", ascending=False, inplace=True)
elif tone == "Angry":
book_recs.sort_values(by="anger", ascending=False, inplace=True)
elif tone == "Suspenseful":
book_recs.sort_values(by="fear", ascending=False, inplace=True)
elif tone == "Sad":
book_recs.sort_values(by="sadness", ascending=False, inplace=True)
return book_recs
def recommend_books(query:str, category:str,tone:str ):
recommendations = retrieve_semantic_recommendations(query, category, tone)
results = []
for _, row in recommendations.iterrows():
description = row["description"]
truncated_desc_split = description.split()
truncated_description = " ".join(truncated_desc_split[:30]) + "..."
authors_split = row["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 = row["authors"]
caption = f"{row['title']} by {authors_str}: {truncated_description}"
results.append((row["large_thumbnail"], caption))
return results
categories = ["All"] + sorted(books["simple_categories"].unique())
tones = ["All"] + ["Happy", "Surprising", "Angry", "Suspenseful", "Sad"]
with gr.Blocks(theme = gr.themes.Glass()) as dashboard:
gr.Markdown("# Semantic book recommender")
with gr.Row():
user_query = gr.Textbox(label = "Please enter a description of a book:",
placeholder = "e.g., A story about forgiveness")
category_dropdown = gr.Dropdown(choices = categories, label = "Select a category:", value = "All")
tone_dropdown = gr.Dropdown(choices = tones, label = "Select an emotional tone:", value = "All")
submit_button = gr.Button("Find recommendations")
gr.Markdown("## Recommendations")
output = gr.Gallery(label = "Recommended books", columns = 8, rows = 2)
submit_button.click(fn = recommend_books,
inputs = [user_query, category_dropdown, tone_dropdown],
outputs = output)
if __name__ == "__main__":
dashboard.launch() |