import streamlit as st
import os
import google.generativeai as genai
from huggingface_hub import hf_hub_download
import base64
from PIL import Image
MODEL_ID = "gemini-2.0-flash-exp" # Keep the model ID as is
try:
api_key = os.getenv("GEMINI_API_KEY")
model_id = MODEL_ID
genai.configure(api_key=api_key)
except Exception as e:
st.error(f"Error: {e}")
st.stop
model = genai.GenerativeModel(MODEL_ID)
chat = model.start_chat()
def download_pdf():
"""
Downloads the PDF file from the Hugging Face Hub using the correct repo path and filename.
"""
try:
hf_token = os.getenv("HF_TOKEN")
repo_id = "wvsuaidev/authorship_verfication_dataset" # Corrected dataset repo path
filename = "Authorship_Verification_Linguistic_Divergence.pdf"
filepath = hf_hub_download(repo_id=repo_id, filename=filename, token=hf_token, repo_type="dataset")
return filepath
except Exception as e:
st.error(f"Failed to download PDF from Hugging Face Hub: {e}")
st.stop() # Stop if the download fails
# Initialize conversation history in Streamlit session state
if "conversation_history" not in st.session_state:
st.session_state.conversation_history = []
if "uploaded_file_part" not in st.session_state: # Store the file *part*
st.session_state.uploaded_file_part = None
if "uploaded_pdf_path" not in st.session_state:
st.session_state.uploaded_pdf_path = download_pdf()
def multimodal_prompt(pdf_path, text_prompt):
"""
Sends a multimodal prompt to Gemini, handling file uploads efficiently.
Args:
pdf_path: The path to the PDF file.
text_prompt: The text prompt for the model.
Returns:
The model's response as a string, or an error message.
"""
try:
if st.session_state.uploaded_file_part is None: # First time, upload
pdf_part = genai.upload_file(pdf_path, mime_type="application/pdf")
st.session_state.uploaded_file_part = pdf_part
prompt = [text_prompt, pdf_part] # First turn includes the actual file
else: # Subsequent turns, reference the file
prompt = [text_prompt, st.session_state.uploaded_file_part] # Subsequent turns include the file reference
response = chat.send_message(prompt)
# Update conversation history
st.session_state.conversation_history.append({"role": "user", "content": text_prompt, "has_pdf": True})
st.session_state.conversation_history.append({"role": "assistant", "content": response.text})
return response.text
except Exception as e:
return f"An error occurred: {e}"
def display_download_button(file_path, file_name):
try:
with open(file_path, "rb") as f:
file_bytes = f.read()
b64 = base64.b64encode(file_bytes).decode()
href = f'<a href="data:application/pdf;base64,{b64}" download="{file_name}">Download the source document (PDF)</a>'
st.markdown(href, unsafe_allow_html=True)
except FileNotFoundError:
st.error("File not found for download.")
except Exception as e:
st.error(f"Error during download: {e}")
# --- Main Page ---
st.title("📚 VQA on the Authorship Attribution and Verification Paper")
about = """
**How to use this App**
This app leverages Gemini 2.0 to provide insights on the provided document.
Select a question from the dropdown menu or enter your own question to get
Gemini's generated response based on the provided document.
"""
with st.expander("How to use this App"):
st.markdown(about)
# --- Load the image ---
image = Image.open("authorship.png")
st.image(image, width=400)
# --- Q and A Tab ---
st.header("Questions and Answers")
# Generate 5 questions based on the selected role
questions = [
"What are the key differences between Authorship Attribution (AA) and Authorship Verification (AV)?",
"What is the 'non-comparability problem' in authorship verification, and why is it significant?",
"How does the proposed DV-Distance metric address the non-comparability problem?",
"Explain the concept of Normal Writing Style (NWS) and its role in the proposed method.",
"How are Deviation Vectors (DVs) calculated, and what do they represent?",
"Describe the two main methods proposed in the paper: DV-Distance and DV-Projection.",
"What are the advantages and limitations of the unsupervised DV-Distance method?",
"How does the supervised DV-Projection method improve upon the DV-Distance method?",
"What language models were used in the study, and why?",
"What datasets were used to evaluate the proposed methods?",
"What evaluation metrics were used in the study?",
"How did the proposed methods perform compared to the baselines and state-of-the-art methods?",
"What were the key findings and trends observed in the experiments?",
"Why did the AWD-LSTM based DV-Distance method consistently outperform the RoBERTa based DV-Distance method?",
"For what types of documents were the proposed methods most suitable?",
"How do the authors explain the performance differences across different document types?",
"What are the potential real-world applications of this research?",
"What are the limitations of the proposed methods?",
"What are some possible directions for future research in this area?",
"How does this research contribute to the broader field of Natural Language Processing (NLP)?"
]
# Create a selection box
selected_question = st.selectbox("Choose a question", questions)
# Display a checkbox
if st.checkbox('Check this box to ask a question not listed above'):
# If the checkbox is checked, display a text box
selected_question = st.text_input('Enter a question')
if st.button("Ask AI"):
with st.spinner("AI is thinking..."):
if st.session_state.uploaded_pdf_path is None:
st.session_state.uploaded_pdf_path = download_pdf()
filepath = st.session_state.uploaded_pdf_path
text_prompt = f"Use the provided document to answer the following question: {selected_question}. Cite the relevant sections of the IRR."
response = multimodal_prompt(filepath, text_prompt) # Use the downloaded filepath
st.markdown(f"**Response:** {response}")
if st.session_state.uploaded_pdf_path:
display_download_button(st.session_state.uploaded_pdf_path, "Visual_Understanding.pdf")
st.markdown("[Visit our Hugging Face Space!](https://huggingface.co/wvsuaidev)")
st.markdown("© 2025 WVSU AI Dev Team 🤖 ✨")