AutoCitation / app.py
yipengsun's picture
Update app.py
05d713a verified
import asyncio
import json
import urllib.parse
import re
import xml.etree.ElementTree as ET
from dataclasses import dataclass, field
from typing import Dict, List, Optional, Any, Tuple
import sys
from loguru import logger
import aiohttp
import gradio as gr
from langchain.prompts import PromptTemplate
from langchain_google_genai import ChatGoogleGenerativeAI
import bibtexparser
from bibtexparser.bwriter import BibTexWriter
from bibtexparser.bibdatabase import BibDatabase
def get_bibtex_writer() -> BibTexWriter:
"""
Create and return a configured BibTexWriter instance.
"""
writer = BibTexWriter()
writer.indent = ' '
writer.comma_first = False
return writer
@dataclass
class Config:
gemini_api_key: str
max_retries: int = 3
base_delay: int = 1
max_queries: int = 5
max_citations_per_query: int = 10
arxiv_base_url: str = 'http://export.arxiv.org/api/query?'
crossref_base_url: str = 'https://api.crossref.org/works'
default_headers: Dict[str, str] = field(default_factory=lambda: {
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36'
})
log_level: str = 'DEBUG'
class ArxivXmlParser:
"""
Class to parse ArXiv XML responses.
"""
NS = {
'atom': 'http://www.w3.org/2005/Atom',
'arxiv': 'http://arxiv.org/schemas/atom'
}
def parse_papers(self, data: str) -> List[Dict[str, Any]]:
"""
Parse ArXiv XML data and return a list of paper dictionaries.
"""
try:
root = ET.fromstring(data)
papers = []
for entry in root.findall('atom:entry', self.NS):
paper = self.parse_entry(entry)
if paper:
papers.append(paper)
return papers
except Exception as e:
logger.error(f"Error parsing ArXiv XML: {e}")
return []
def parse_entry(self, entry: ET.Element) -> Optional[Dict[str, Any]]:
"""
Parse a single ArXiv entry element and return a dictionary with paper details.
"""
try:
title_node = entry.find('atom:title', self.NS)
if title_node is None:
return None
title = title_node.text.strip() if title_node.text else ""
authors = []
for author in entry.findall('atom:author', self.NS):
author_name_node = author.find('atom:name', self.NS)
if author_name_node is not None and author_name_node.text:
authors.append(self._format_author_name(author_name_node.text.strip()))
arxiv_id_node = entry.find('atom:id', self.NS)
if arxiv_id_node is None or not arxiv_id_node.text:
return None
arxiv_id = arxiv_id_node.text.split('/')[-1]
published_node = entry.find('atom:published', self.NS)
year = published_node.text[:4] if (published_node is not None and published_node.text) else "Unknown"
abstract_node = entry.find('atom:summary', self.NS)
abstract = abstract_node.text.strip() if (abstract_node is not None and abstract_node.text) else ""
bibtex_key = f"{authors[0].split(',')[0]}{arxiv_id.replace('.', '')}" if authors else f"unknown{arxiv_id.replace('.', '')}"
bibtex_entry = self._generate_bibtex_entry(bibtex_key, title, authors, arxiv_id, year)
return {
'title': title,
'authors': authors,
'arxiv_id': arxiv_id,
'published': year,
'abstract': abstract,
'bibtex_key': bibtex_key,
'bibtex_entry': bibtex_entry
}
except Exception as e:
logger.error(f"Error parsing ArXiv entry: {e}")
return None
@staticmethod
def _format_author_name(author: str) -> str:
"""
Format an author name as 'Lastname, Firstname'.
"""
names = author.split()
if len(names) > 1:
return f"{names[-1]}, {' '.join(names[:-1])}"
return author
def _generate_bibtex_entry(self, key: str, title: str, authors: List[str], arxiv_id: str, year: str) -> str:
"""
Generate a BibTeX entry for a paper.
"""
db = BibDatabase()
db.entries = [{
'ENTRYTYPE': 'article',
'ID': key,
'title': title,
'author': ' and '.join(authors),
'journal': f'arXiv preprint arXiv:{arxiv_id}',
'year': year
}]
writer = get_bibtex_writer()
return writer.write(db).strip()
class AsyncContextManager:
"""
Asynchronous context manager to handle aiohttp ClientSession.
"""
async def __aenter__(self) -> aiohttp.ClientSession:
self._session = aiohttp.ClientSession()
return self._session
async def __aexit__(self, *_):
if self._session:
await self._session.close()
class CitationGenerator:
"""
Class that handles generating citations using AI and searching for academic papers.
"""
def __init__(self, config: Config) -> None:
self.config = config
self.xml_parser = ArxivXmlParser()
self.async_context = AsyncContextManager()
self.llm = ChatGoogleGenerativeAI(
model="gemini-2.0-flash",
temperature=0.3,
google_api_key=config.gemini_api_key,
streaming=True
)
self.citation_prompt = PromptTemplate.from_template(
"""Insert citations into the provided text using LaTeX \\cite{{key}} commands.
You must not alter the original wording or structure of the text beyond adding citations.
You must include all provided references at least once. Place citations at suitable points.
Input text:
{text}
Available papers (cite each at least once):
{papers}
"""
)
self.generate_queries_prompt = PromptTemplate.from_template(
"""Generate {num_queries} diverse academic search queries based on the given text.
The queries should be concise and relevant.
Requirements:
1. Return ONLY a valid JSON array of strings.
2. No additional text or formatting beyond JSON.
3. Ensure uniqueness.
Text: {text}
"""
)
logger.remove()
logger.add(sys.stderr, level=config.log_level)
async def generate_queries(self, text: str, num_queries: int) -> List[str]:
"""
Generate a list of academic search queries from the input text.
"""
input_map = {
"text": text,
"num_queries": num_queries
}
try:
prompt = self.generate_queries_prompt.format(**input_map)
response = await self.llm.ainvoke(prompt)
content = response.content.strip()
if not content.startswith('['):
start = content.find('[')
end = content.rfind(']') + 1
if start >= 0 and end > start:
content = content[start:end]
try:
queries = json.loads(content)
if isinstance(queries, list):
return [q.strip() for q in queries if isinstance(q, str)][:num_queries]
except json.JSONDecodeError:
lines = [line.strip() for line in content.split('\n')
if line.strip() and not line.strip().startswith(('[', ']'))]
return lines[:num_queries]
return ["deep learning neural networks"]
except Exception as e:
logger.error(f"Error generating queries: {e}")
return ["deep learning neural networks"]
async def search_arxiv(self, session: aiohttp.ClientSession, query: str, max_results: int) -> List[Dict[str, Any]]:
"""
Search ArXiv for papers matching the query.
"""
try:
params = {
'search_query': f'all:{urllib.parse.quote(query)}',
'start': 0,
'max_results': max_results,
'sortBy': 'relevance',
'sortOrder': 'descending'
}
url = self.config.arxiv_base_url + urllib.parse.urlencode(params)
async with session.get(
url,
headers=self.config.default_headers,
timeout=30
) as response:
text_data = await response.text()
papers = self.xml_parser.parse_papers(text_data)
return papers
except Exception as e:
logger.error(f"Error searching ArXiv: {e}")
return []
async def fix_author_name(self, author: str) -> str:
"""
Correct an author name that contains corrupted characters.
"""
if not re.search(r'[οΏ½]', author):
return author
try:
prompt = f"""Fix this author name that contains corrupted characters (οΏ½):
Name: {author}
Requirements:
1. Return ONLY the fixed author name
2. Use proper diacritical marks for names
3. Consider common name patterns and languages
4. If unsure, use the most likely letter
5. Maintain the format: "Lastname, Firstname"
"""
response = await self.llm.ainvoke(prompt)
fixed_name = response.content.strip()
return fixed_name if fixed_name else author
except Exception as e:
logger.error(f"Error fixing author name: {e}")
return author
async def format_bibtex_author_names(self, text: str) -> str:
"""
Clean and format author names in a BibTeX string.
"""
try:
bib_database = bibtexparser.loads(text)
for entry in bib_database.entries:
if 'author' in entry:
authors = entry['author'].split(' and ')
cleaned_authors = []
for author in authors:
fixed_author = await self.fix_author_name(author)
cleaned_authors.append(fixed_author)
entry['author'] = ' and '.join(cleaned_authors)
writer = get_bibtex_writer()
return writer.write(bib_database).strip()
except Exception as e:
logger.error(f"Error cleaning BibTeX special characters: {e}")
return text
async def search_crossref(self, session: aiohttp.ClientSession, query: str, max_results: int) -> List[Dict[str, Any]]:
"""
Search CrossRef for papers matching the query.
"""
try:
cleaned_query = query.replace("'", "").replace('"', "")
if ' ' in cleaned_query:
cleaned_query = f'"{cleaned_query}"'
params = {
'query.bibliographic': cleaned_query,
'rows': max_results,
'select': 'DOI,title,author,published-print,container-title',
'sort': 'relevance',
'order': 'desc'
}
headers = {
'User-Agent': 'Mozilla/5.0 (compatible; CitationBot/1.0; mailto:[email protected])',
'Accept': 'application/json'
}
for attempt in range(self.config.max_retries):
try:
async with session.get(
self.config.crossref_base_url,
params=params,
headers=headers,
timeout=30
) as response:
if response.status == 429:
delay = self.config.base_delay * (2 ** attempt)
logger.warning(f"Rate limited by CrossRef. Retrying in {delay} seconds...")
await asyncio.sleep(delay)
continue
response.raise_for_status()
search_data = await response.json()
items = search_data.get('message', {}).get('items', [])
if not items:
return []
papers = []
existing_keys = set()
for item in items:
doi = item.get('DOI')
if not doi:
continue
try:
bibtex_url = f"https://doi.org/{doi}"
async with session.get(
bibtex_url,
headers={
'Accept': 'application/x-bibtex',
'User-Agent': 'Mozilla/5.0 (compatible; CitationBot/1.0; mailto:[email protected])'
},
timeout=30
) as bibtex_response:
if bibtex_response.status != 200:
continue
bibtex_text = await bibtex_response.text()
bib_database = bibtexparser.loads(bibtex_text)
if not bib_database.entries:
continue
entry = bib_database.entries[0]
if 'title' not in entry and 'booktitle' not in entry:
continue
if 'author' not in entry:
continue
title = entry.get('title', 'No Title').replace('{', '').replace('}', '')
authors = entry.get('author', 'Unknown').replace('\n', ' ').replace('\t', ' ').strip()
year = entry.get('year', 'Unknown')
key = self._generate_unique_bibtex_key(entry, existing_keys)
entry['ID'] = key
existing_keys.add(key)
writer = get_bibtex_writer()
formatted_bibtex = writer.write(bib_database).strip()
papers.append({
'title': title,
'authors': authors,
'year': year,
'bibtex_key': key,
'bibtex_entry': formatted_bibtex
})
except Exception as e:
logger.error(f"Error processing CrossRef item: {e}")
return papers
except aiohttp.ClientError as e:
if attempt == self.config.max_retries - 1:
logger.error(f"Max retries reached for CrossRef search. Error: {e}")
raise
delay = self.config.base_delay * (2 ** attempt)
logger.warning(f"Client error during CrossRef search: {e}. Retrying in {delay} seconds...")
await asyncio.sleep(delay)
except Exception as e:
logger.error(f"Error searching CrossRef: {e}")
return []
def _generate_unique_bibtex_key(self, entry: Dict[str, Any], existing_keys: set) -> str:
"""
Generate a unique BibTeX key for an entry.
"""
entry_type = entry.get('ENTRYTYPE', '').lower()
author_field = entry.get('author', '')
year = entry.get('year', '')
authors = [a.strip() for a in author_field.split(' and ')]
first_author_last_name = authors[0].split(',')[0] if authors else 'unknown'
if entry_type == 'inbook':
booktitle = entry.get('booktitle', '')
title_word = re.sub(r'\W+', '', booktitle.split()[0]) if booktitle.split() else 'untitled'
else:
title = entry.get('title', '')
title_word = re.sub(r'\W+', '', title.split()[0]) if title.split() else 'untitled'
base_key = f"{first_author_last_name}{year}{title_word}"
key = base_key
index = 1
while key in existing_keys:
key = f"{base_key}{index}"
index += 1
return key
async def process_text(self, text: str, num_queries: int, citations_per_query: int,
use_arxiv: bool = True, use_crossref: bool = True) -> Tuple[str, str, str]:
"""
Process the input text to generate citations and corresponding BibTeX entries.
"""
if not (use_arxiv or use_crossref):
return "Please select at least one source (ArXiv or CrossRef)", "", ""
num_queries = min(max(1, num_queries), self.config.max_queries)
citations_per_query = min(max(1, citations_per_query), self.config.max_citations_per_query)
async def generate_queries_tool(input_data: Dict[str, Any]) -> List[str]:
return await self.generate_queries(input_data["text"], input_data["num_queries"])
async def search_papers_tool(input_data: Dict[str, Any]) -> List[Dict[str, Any]]:
queries = input_data["queries"]
papers = []
async with self.async_context as session:
search_tasks = []
for q in queries:
if input_data["use_arxiv"]:
search_tasks.append(self.search_arxiv(session, q, input_data["citations_per_query"]))
if input_data["use_crossref"]:
search_tasks.append(self.search_crossref(session, q, input_data["citations_per_query"]))
results = await asyncio.gather(*search_tasks, return_exceptions=True)
for r in results:
if not isinstance(r, Exception):
papers.extend(r)
# Remove duplicate papers
unique_papers = []
seen_keys = set()
for p in papers:
if p['bibtex_key'] not in seen_keys:
seen_keys.add(p['bibtex_key'])
unique_papers.append(p)
return unique_papers
async def cite_text_tool(input_data: Dict[str, Any]) -> Tuple[str, str]:
try:
citation_input = {
"text": input_data["text"],
"papers": json.dumps(input_data["papers"], indent=2)
}
prompt = self.citation_prompt.format(**citation_input)
response = await self.llm.ainvoke(prompt)
cited_text = response.content.strip()
bib_database = BibDatabase()
for p in input_data["papers"]:
if 'bibtex_entry' in p:
bib_db = bibtexparser.loads(p['bibtex_entry'])
if bib_db.entries:
bib_database.entries.append(bib_db.entries[0])
else:
logger.warning(f"Empty BibTeX entry for key: {p['bibtex_key']}")
writer = get_bibtex_writer()
bibtex_entries = writer.write(bib_database).strip()
return cited_text, bibtex_entries
except Exception as e:
logger.error(f"Error inserting citations: {e}")
return input_data["text"], ""
async def agent_run(input_data: Dict[str, Any]) -> Tuple[str, str, str]:
queries = await generate_queries_tool(input_data)
papers = await search_papers_tool({
"queries": queries,
"citations_per_query": input_data["citations_per_query"],
"use_arxiv": input_data["use_arxiv"],
"use_crossref": input_data["use_crossref"]
})
if not papers:
return input_data["text"], "", "\n".join([f"- {q}" for q in queries])
cited_text, final_bibtex = await cite_text_tool({
"text": input_data["text"],
"papers": papers
})
return cited_text, final_bibtex, "\n".join([f"- {q}" for q in queries])
final_text, final_bibtex, final_queries = await agent_run({
"text": text,
"num_queries": num_queries,
"citations_per_query": citations_per_query,
"use_arxiv": use_arxiv,
"use_crossref": use_crossref
})
return final_text, final_bibtex, final_queries
def create_gradio_interface() -> gr.Interface:
"""
Create and return a Gradio interface for the citation generator.
"""
async def process(api_key: str, text: str, num_queries: int, citations_per_query: int,
use_arxiv: bool, use_crossref: bool) -> Tuple[str, str, str]:
if not api_key.strip():
return "Please enter your Gemini API Key.", "", ""
if not text.strip():
return "Please enter text to process", "", ""
try:
config = Config(gemini_api_key=api_key)
citation_gen = CitationGenerator(config)
return await citation_gen.process_text(
text, num_queries, citations_per_query,
use_arxiv=use_arxiv, use_crossref=use_crossref
)
except ValueError as e:
return f"Input validation error: {str(e)}", "", ""
except Exception as e:
return f"Error: {str(e)}", "", ""
css = """
:root {
/* Modern color palette */
--primary-bg: #F8F9FA;
--secondary-bg: #FFFFFF;
--accent-1: #4A90E2;
--accent-2: #50C878;
--accent-3: #F5B041;
--text-primary: #2C3E50;
--text-secondary: #566573;
--border: #E5E7E9;
--shadow: rgba(0, 0, 0, 0.1);
}
body {
background-color: var(--primary-bg);
color: var(--text-primary);
font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Oxygen, Ubuntu, Cantarell, sans-serif;
line-height: 1.6;
}
.header {
text-align: center;
margin-bottom: 2rem;
padding: 1.5rem;
background-color: var(--secondary-bg);
box-shadow: 0 2px 4px var(--shadow);
border-bottom: none;
}
.header h1 {
font-size: 2.25rem;
color: var(--accent-1);
margin-bottom: 0.75rem;
font-weight: 600;
}
.header p {
font-size: 1.1rem;
color: var(--text-secondary);
}
.input-group, .controls-row, .source-controls, .output-group {
padding: 1rem;
margin-bottom: 1.5rem;
border: 1px solid var(--border);
border-radius: 12px;
background-color: var(--secondary-bg);
box-shadow: 0 1px 3px var(--shadow);
}
.input-group label, .controls-row label, .source-controls label {
color: var(--text-primary);
font-weight: 500;
margin-bottom: 0.5rem;
display: block;
}
input[type="number"], textarea, .gradio-input, .gradio-output {
border: 1px solid var(--border);
border-radius: 8px;
padding: 0.75rem;
background-color: var(--primary-bg);
color: var(--text-primary);
font-size: 1rem;
width: 100%;
transition: border-color 0.3s, box-shadow 0.3s;
}
input[type="number"]:focus, textarea:focus {
border-color: var(--accent-1);
box-shadow: 0 0 0 2px rgba(74, 144, 226, 0.1);
outline: none;
}
.generate-btn {
background-color: var(--accent-1);
color: white;
padding: 1rem 2rem;
border: none;
border-radius: 8px;
font-size: 1.1rem;
font-weight: 500;
cursor: pointer;
transition: all 0.3s ease;
width: 100%;
}
.generate-btn:hover {
background-color: #357ABD;
transform: translateY(-1px);
box-shadow: 0 4px 6px var(--shadow);
}
.gradio-button {
background-color: var(--accent-2) !important;
border-radius: 8px !important;
transition: all 0.3s ease !important;
}
.gradio-button:hover {
background-color: #45B76C !important;
transform: translateY(-1px);
}
.gradio-copy-button {
background-color: var(--accent-3) !important;
color: var(--text-primary) !important;
border: none !important;
border-radius: 6px !important;
padding: 0.4rem 0.8rem !important;
cursor: pointer !important;
font-size: 0.9rem !important;
font-weight: 500 !important;
transition: all 0.3s ease !important;
}
.gradio-copy-button:hover {
background-color: #F39C12 !important;
transform: translateY(-1px);
box-shadow: 0 2px 4px var(--shadow);
}
"""
with gr.Blocks(css=css, theme=gr.themes.Default()) as demo:
gr.HTML("""
<div class="header">
<h1>πŸ“š AutoCitation</h1>
<p>An AI agent that automatically adds citations into your academic text</p>
</div>
""")
api_key = gr.Textbox(
label="Gemini API Key",
placeholder="Enter your Gemini API key...",
type="password"
)
input_text = gr.Textbox(
label="Input Text",
placeholder="Paste or type your text here...",
lines=8
)
with gr.Row():
num_queries = gr.Number(
label="Search Queries",
value=3,
minimum=1,
maximum=Config.max_queries,
step=1
)
citations_per_query = gr.Number(
label="Citations per Query",
value=1,
minimum=1,
maximum=Config.max_citations_per_query,
step=1
)
with gr.Row():
use_arxiv = gr.Checkbox(
label="Search arXiv",
value=True
)
use_crossref = gr.Checkbox(
label="Search Crossref",
value=True
)
process_btn = gr.Button("Generate", elem_classes="generate-btn")
with gr.Row():
cited_text = gr.Textbox(
label="Generated Text",
lines=10,
show_copy_button=True
)
bibtex = gr.Textbox(
label="BibTeX References",
lines=10,
show_copy_button=True
)
queries_text = gr.Textbox(
label="Generated Queries",
lines=10,
show_copy_button=True
)
process_btn.click(
fn=process,
inputs=[api_key, input_text, num_queries, citations_per_query, use_arxiv, use_crossref],
outputs=[cited_text, bibtex, queries_text]
)
return demo
if __name__ == "__main__":
demo = create_gradio_interface()
try:
demo.launch(server_port=7860, share=False)
except KeyboardInterrupt:
print("\nShutting down server...")
except Exception as e:
print(f"Error starting server: {str(e)}")