academic-refchecker 2.0.7__py3-none-any.whl

refchecker/checkers/openreview_checker.py

@@ -0,0 +1,984 @@
+#!/usr/bin/env python3
+"""
+OpenReview API Client for Reference Verification
+
+This module provides functionality to verify references from OpenReview papers.
+OpenReview is a platform for open peer review in machine learning conferences
+like ICLR, NeurIPS, ICML, etc.
+
+Usage:
+    from openreview_checker import OpenReviewReferenceChecker
+    
+    # Initialize the checker
+    checker = OpenReviewReferenceChecker()
+    
+    # Verify a reference
+    reference = {
+        'title': 'Title of the paper',
+        'authors': ['Author 1', 'Author 2'],
+        'year': 2024,
+        'url': 'https://openreview.net/forum?id=ZG3RaNIsO8',
+        'raw_text': 'Full citation text'
+    }
+    
+    verified_data, errors, url = checker.verify_reference(reference)
+"""
+
+import requests
+import time
+import logging
+import re
+import json
+from typing import Dict, List, Tuple, Optional, Any, Union
+from urllib.parse import urlparse, parse_qs
+from bs4 import BeautifulSoup
+from refchecker.utils.text_utils import (
+    normalize_text, clean_title_basic, is_name_match, 
+    calculate_title_similarity, compare_authors, 
+    clean_title_for_search, are_venues_substantially_different,
+    is_year_substantially_different, strip_latex_commands,
+    compare_titles_with_latex_cleaning
+)
+
+# Set up logging
+logger = logging.getLogger(__name__)
+
+class OpenReviewReferenceChecker:
+    """
+    A class to verify references using OpenReview
+    """
+    
+    def __init__(self, request_delay: float = 1.0):
+        """
+        Initialize the OpenReview client
+        
+        Args:
+            request_delay: Delay between requests to be respectful to OpenReview servers
+        """
+        self.base_url = "https://openreview.net"
+        self.api_url = "https://api.openreview.net"
+        self.request_delay = request_delay
+        self.last_request_time = 0
+        
+        # Session for connection pooling
+        self.session = requests.Session()
+        self.session.headers.update({
+            'User-Agent': 'RefChecker/1.0 (Academic Reference Verification)',
+            'Accept': 'application/json, text/html',
+            'Accept-Language': 'en-US,en;q=0.9'
+        })
+    
+    def is_openreview_url(self, url: str) -> bool:
+        """
+        Check if URL is from OpenReview
+        
+        Args:
+            url: URL to check
+            
+        Returns:
+            True if it's an OpenReview URL
+        """
+        return bool(url and 'openreview.net' in url.lower())
+    
+    def is_openreview_reference(self, reference: Dict[str, Any]) -> bool:
+        """
+        Determine if this reference is from OpenReview based on URL patterns
+        
+        Args:
+            reference: Reference dictionary to check
+            
+        Returns:
+            True if reference appears to be from OpenReview
+        """
+        # Check various URL fields for OpenReview URLs
+        url_fields = ['url', 'openreview_url', 'link', 'venue_url']
+        for field in url_fields:
+            url = reference.get(field, '')
+            if url and self.is_openreview_url(url):
+                return True
+        
+        # Check raw text for OpenReview URLs  
+        raw_text = reference.get('raw_text', '')
+        if raw_text and 'openreview.net' in raw_text.lower():
+            return True
+            
+        return False
+    
+    def extract_paper_id(self, url: str) -> Optional[str]:
+        """
+        Extract paper ID from OpenReview URL
+        
+        Args:
+            url: OpenReview URL
+            
+        Returns:
+            Paper ID if found, None otherwise
+        """
+        if not self.is_openreview_url(url):
+            return None
+        
+        # Handle different OpenReview URL formats:
+        # https://openreview.net/forum?id=ZG3RaNIsO8
+        # https://openreview.net/pdf?id=ZG3RaNIsO8
+        # https://openreview.net/forum?id=ZG3RaNIsO8&noteId=...
+        
+        parsed = urlparse(url)
+        query_params = parse_qs(parsed.query)
+        
+        if 'id' in query_params:
+            return query_params['id'][0]
+        
+        # Also check path-based URLs (if they exist)
+        path_match = re.search(r'/(?:forum|pdf|notes)/([A-Za-z0-9_-]+)', parsed.path)
+        if path_match:
+            return path_match.group(1)
+        
+        return None
+    
+    def _respectful_request(self, url: str, **kwargs) -> Optional[requests.Response]:
+        """Make a respectful HTTP request with rate limiting"""
+        current_time = time.time()
+        time_since_last = current_time - self.last_request_time
+        
+        if time_since_last < self.request_delay:
+            time.sleep(self.request_delay - time_since_last)
+        
+        try:
+            logger.debug(f"Making request to: {url}")
+            response = self.session.get(url, timeout=15, **kwargs)
+            self.last_request_time = time.time()
+            logger.debug(f"Request successful: {response.status_code}")
+            return response
+        except requests.exceptions.RequestException as e:
+            logger.debug(f"Request failed for {url}: {type(e).__name__}: {e}")
+            return None
+    
+    def get_paper_metadata(self, paper_id: str) -> Optional[Dict[str, Any]]:
+        """
+        Get paper metadata from OpenReview
+        
+        Args:
+            paper_id: OpenReview paper ID
+            
+        Returns:
+            Paper metadata dictionary or None if not found
+        """
+        # Try API endpoint first
+        api_url = f"{self.api_url}/notes?id={paper_id}"
+        response = self._respectful_request(api_url)
+        
+        if response and response.status_code == 200:
+            try:
+                data = response.json()
+                if 'notes' in data and data['notes']:
+                    note = data['notes'][0]
+                    return self._parse_api_response(note)
+            except (json.JSONDecodeError, KeyError) as e:
+                logger.debug(f"Failed to parse API response: {e}")
+        
+        # Fall back to web scraping
+        forum_url = f"{self.base_url}/forum?id={paper_id}"
+        response = self._respectful_request(forum_url)
+        
+        if not response or response.status_code != 200:
+            return None
+        
+        return self._parse_web_page(response.text, forum_url)
+    
+    def _parse_api_response(self, note: Dict[str, Any]) -> Dict[str, Any]:
+        """
+        Parse OpenReview API response to extract metadata
+        
+        Args:
+            note: Note data from API response
+            
+        Returns:
+            Parsed metadata dictionary
+        """
+        content = note.get('content', {})
+        
+        # Extract basic metadata
+        metadata = {
+            'id': note.get('id'),
+            'title': content.get('title', '').strip(),
+            'authors': [],
+            'year': None,
+            'venue': None,
+            'abstract': content.get('abstract', '').strip(),
+            'keywords': content.get('keywords', []),
+            'pdf_url': content.get('pdf'),
+            'forum_url': f"{self.base_url}/forum?id={note.get('id')}",
+            'source': 'openreview_api'
+        }
+        
+        # Parse authors
+        authors_raw = content.get('authors', [])
+        if isinstance(authors_raw, list):
+            metadata['authors'] = [author.strip() for author in authors_raw if author.strip()]
+        elif isinstance(authors_raw, str):
+            # Sometimes authors are in a single string
+            metadata['authors'] = [author.strip() for author in authors_raw.split(',') if author.strip()]
+        
+        # Extract year from various sources
+        # Check creation time
+        if 'cdate' in note:
+            try:
+                import datetime
+                timestamp = note['cdate'] / 1000.0  # Convert from milliseconds
+                year = datetime.datetime.fromtimestamp(timestamp).year
+                metadata['year'] = year
+            except (ValueError, TypeError):
+                pass
+        
+        # Check if venue/conference info is available
+        venue_info = content.get('venue', '')
+        if venue_info:
+            metadata['venue'] = venue_info.strip()
+        
+        # Try to extract venue from forum context or submission info
+        if not metadata['venue']:
+            # Common venues for OpenReview
+            forum_path = note.get('forum', '')
+            if 'ICLR' in str(content) or 'iclr' in forum_path.lower():
+                metadata['venue'] = 'ICLR'
+            elif 'NeurIPS' in str(content) or 'neurips' in forum_path.lower():
+                metadata['venue'] = 'NeurIPS'
+            elif 'ICML' in str(content) or 'icml' in forum_path.lower():
+                metadata['venue'] = 'ICML'
+        
+        return metadata
+    
+    def _parse_web_page(self, html: str, url: str) -> Dict[str, Any]:
+        """
+        Parse OpenReview web page to extract metadata
+        
+        Args:
+            html: HTML content of the page
+            url: Original URL
+            
+        Returns:
+            Parsed metadata dictionary
+        """
+        soup = BeautifulSoup(html, 'html.parser')
+        
+        # Extract paper ID from URL
+        paper_id = self.extract_paper_id(url)
+        
+        metadata = {
+            'id': paper_id,
+            'title': '',
+            'authors': [],
+            'year': None,
+            'venue': None,
+            'abstract': '',
+            'keywords': [],
+            'forum_url': url,
+            'source': 'openreview_web'
+        }
+        
+        # Extract title
+        title_elem = soup.find('h2', {'class': 'citation_title'}) or soup.find('h1')
+        if title_elem:
+            metadata['title'] = title_elem.get_text().strip()
+        
+        # Try to find title in meta tags
+        if not metadata['title']:
+            meta_title = soup.find('meta', {'property': 'og:title'}) or soup.find('meta', {'name': 'title'})
+            if meta_title and meta_title.get('content'):
+                metadata['title'] = meta_title['content'].strip()
+        
+        # Extract authors from meta tags (most reliable for OpenReview)
+        author_metas = soup.find_all('meta', {'name': 'citation_author'})
+        if author_metas:
+            metadata['authors'] = [meta.get('content', '').strip() for meta in author_metas if meta.get('content', '').strip()]
+        
+        # Fallback: try to find authors in HTML structure
+        if not metadata['authors']:
+            authors_section = soup.find('div', {'class': 'authors'}) or soup.find('span', {'class': 'authors'})
+            if authors_section:
+                # Extract author names from links or text
+                author_links = authors_section.find_all('a')
+                if author_links:
+                    metadata['authors'] = [link.get_text().strip() for link in author_links]
+                else:
+                    # Parse comma-separated authors
+                    authors_text = authors_section.get_text().strip()
+                    metadata['authors'] = [author.strip() for author in authors_text.split(',') if author.strip()]
+        
+        # Extract year from various sources
+        year_pattern = r'\b(20\d{2})\b'
+        
+        # Check date/year elements
+        date_elem = soup.find('span', {'class': 'date'}) or soup.find('time')
+        if date_elem:
+            year_match = re.search(year_pattern, date_elem.get_text())
+            if year_match:
+                metadata['year'] = int(year_match.group(1))
+        
+        # Check meta tags for date
+        if not metadata['year']:
+            meta_date = soup.find('meta', {'name': 'citation_date'}) or soup.find('meta', {'name': 'date'})
+            if meta_date and meta_date.get('content'):
+                year_match = re.search(year_pattern, meta_date['content'])
+                if year_match:
+                    metadata['year'] = int(year_match.group(1))
+        
+        # Extract abstract
+        abstract_elem = soup.find('div', {'class': 'abstract'}) or soup.find('section', {'class': 'abstract'})
+        if abstract_elem:
+            metadata['abstract'] = abstract_elem.get_text().strip()
+        
+        # Extract venue information from meta tags (most reliable for OpenReview)
+        venue_meta = soup.find('meta', {'name': 'citation_conference_title'})
+        if venue_meta and venue_meta.get('content'):
+            venue_full = venue_meta['content'].strip()
+            # Convert long conference names to common abbreviations
+            if 'International Conference on Learning Representations' in venue_full:
+                # Extract year if present
+                year_match = re.search(r'\b(20\d{2})\b', venue_full)
+                if year_match:
+                    metadata['venue'] = f'ICLR {year_match.group(1)}'
+                else:
+                    metadata['venue'] = 'ICLR'
+            elif 'Neural Information Processing Systems' in venue_full or 'NeurIPS' in venue_full:
+                year_match = re.search(r'\b(20\d{2})\b', venue_full)
+                if year_match:
+                    metadata['venue'] = f'NeurIPS {year_match.group(1)}'
+                else:
+                    metadata['venue'] = 'NeurIPS'
+            else:
+                metadata['venue'] = venue_full
+        
+        # Fallback: try HTML structure  
+        if not metadata['venue']:
+            venue_elem = soup.find('div', {'class': 'venue'}) or soup.find('span', {'class': 'venue'})
+            if venue_elem:
+                metadata['venue'] = venue_elem.get_text().strip()
+        
+        # Final fallback: try to determine venue from page context or URL
+        if not metadata['venue']:
+            page_text = soup.get_text().lower()
+            if 'iclr' in page_text or 'iclr' in url.lower():
+                if '2024' in page_text:
+                    metadata['venue'] = 'ICLR 2024'
+                else:
+                    metadata['venue'] = 'ICLR'
+            elif 'neurips' in page_text or 'neurips' in url.lower():
+                metadata['venue'] = 'NeurIPS'
+            elif 'icml' in page_text or 'icml' in url.lower():
+                metadata['venue'] = 'ICML'
+        
+        # Extract keywords if available
+        keywords_elem = soup.find('div', {'class': 'keywords'})
+        if keywords_elem:
+            keywords_text = keywords_elem.get_text()
+            metadata['keywords'] = [kw.strip() for kw in keywords_text.split(',') if kw.strip()]
+        
+        return metadata
+    
+    def verify_reference(self, reference: Dict[str, Any]) -> Tuple[Optional[Dict[str, Any]], List[Dict[str, Any]], Optional[str]]:
+        """
+        Verify a reference against OpenReview
+        
+        Args:
+            reference: Reference dictionary with title, authors, year, url, etc.
+            
+        Returns:
+            Tuple of (verified_data, errors, paper_url) where:
+            - verified_data: Dict with verified OpenReview paper data or None
+            - errors: List of error/warning dictionaries
+            - paper_url: The OpenReview URL
+        """
+        logger.debug(f"Verifying OpenReview reference: {reference.get('title', 'Untitled')}")
+        
+        # Extract OpenReview URL from reference
+        openreview_url = None
+        for url_key in ['url', 'openreview_url', 'link']:
+            if url_key in reference and reference[url_key]:
+                url = reference[url_key].strip()
+                if self.is_openreview_url(url):
+                    openreview_url = url
+                    break
+        
+        if not openreview_url:
+            logger.debug("No OpenReview URL found in reference")
+            return None, [], None
+        
+        # Extract paper ID
+        paper_id = self.extract_paper_id(openreview_url)
+        if not paper_id:
+            return None, [{"error_type": "unverified", "error_details": "Could not extract paper ID from OpenReview URL"}], openreview_url
+        
+        # Get paper metadata
+        paper_data = self.get_paper_metadata(paper_id)
+        if not paper_data:
+            return None, [{"error_type": "unverified", "error_details": "Paper not found on OpenReview"}], openreview_url
+        
+        logger.debug(f"Found OpenReview paper: {paper_data.get('title', 'Untitled')}")
+        
+        # Verify the reference against the paper data
+        errors = []
+        
+        # Check title match
+        cited_title = reference.get('title', '').strip()
+        paper_title = paper_data.get('title', '').strip()
+        
+        if cited_title and paper_title:
+            similarity = compare_titles_with_latex_cleaning(cited_title, paper_title)
+            if similarity < 0.7:  # Using a reasonable threshold
+                from refchecker.utils.error_utils import format_title_mismatch
+                # Clean the cited title for display (remove LaTeX commands like {LLM}s -> LLMs)
+                clean_cited_title = strip_latex_commands(cited_title)
+                details = format_title_mismatch(clean_cited_title, paper_title) + f" (similarity: {similarity:.2f})"
+                errors.append({
+                    "warning_type": "title",
+                    "warning_details": details
+                })
+        
+        # Check authors
+        cited_authors = reference.get('authors', [])
+        paper_authors = paper_data.get('authors', [])
+        
+        if cited_authors and paper_authors:
+            # Convert to list format if needed
+            if isinstance(cited_authors, str):
+                cited_authors = [author.strip() for author in cited_authors.split(',')]
+            if isinstance(paper_authors, str):
+                paper_authors = [author.strip() for author in paper_authors.split(',')]
+            
+            # Use the existing author comparison function
+            match, error_msg = compare_authors(cited_authors, paper_authors)
+            if not match and error_msg:
+                errors.append({
+                    "warning_type": "author",
+                    "warning_details": error_msg
+                })
+        
+        # Check year
+        cited_year = reference.get('year')
+        paper_year = paper_data.get('year')
+        
+        if cited_year and paper_year:
+            try:
+                cited_year_int = int(cited_year)
+                paper_year_int = int(paper_year)
+                
+                is_different, year_message = is_year_substantially_different(cited_year_int, paper_year_int)
+                if is_different and year_message:
+                    from refchecker.utils.error_utils import format_year_mismatch
+                    errors.append({
+                        "warning_type": "year",
+                        "warning_details": format_year_mismatch(cited_year_int, paper_year_int)
+                    })
+            except (ValueError, TypeError):
+                pass  # Skip year validation if conversion fails
+        
+        # Check venue if provided in reference
+        cited_venue = reference.get('venue', '').strip()
+        paper_venue = paper_data.get('venue', '').strip()
+        
+        if cited_venue and paper_venue:
+            if are_venues_substantially_different(cited_venue, paper_venue):
+                from refchecker.utils.error_utils import format_venue_mismatch
+                errors.append({
+                    "warning_type": "venue",
+                    "warning_details": format_venue_mismatch(cited_venue, paper_venue)
+                })
+        
+        # Create verified data structure
+        verified_data = {
+            'title': paper_data.get('title', cited_title),
+            'authors': paper_data.get('authors', cited_authors),
+            'year': paper_data.get('year', cited_year),
+            'venue': paper_data.get('venue', cited_venue),
+            'url': openreview_url,
+            'abstract': paper_data.get('abstract', ''),
+            'keywords': paper_data.get('keywords', []),
+            'openreview_metadata': paper_data,
+            'verification_source': 'OpenReview'
+        }
+        
+        logger.debug(f"OpenReview verification completed for: {openreview_url}")
+        return verified_data, errors, openreview_url
+    
+    def verify_by_search(self, reference: Dict[str, Any]) -> Tuple[Optional[Dict[str, Any]], List[Dict[str, Any]], Optional[str]]:
+        """
+        Verify a reference by searching OpenReview (when no URL is provided)
+        
+        Args:
+            reference: Reference dictionary with title, authors, year, etc.
+            
+        Returns:
+            Tuple of (verified_data, errors, paper_url) where:
+            - verified_data: Dict with verified OpenReview paper data or None
+            - errors: List of error/warning dictionaries  
+            - paper_url: The OpenReview URL if found
+        """
+        logger.debug(f"Searching OpenReview for reference: {reference.get('title', 'Untitled')}")
+        
+        title = reference.get('title', '').strip()
+        authors = reference.get('authors', [])
+        year = reference.get('year')
+        venue = reference.get('venue', '').strip()
+        
+        if not title:
+            return None, [], None
+        
+        # Check if venue suggests this might be on OpenReview
+        if not self._is_likely_openreview_venue(venue):
+            logger.debug(f"Venue '{venue}' doesn't suggest OpenReview, skipping search")
+            return None, [], None
+        
+        # Search for matching papers
+        search_results = self.search_paper(title, authors, year)
+        
+        if not search_results:
+            logger.debug("No matching papers found on OpenReview")
+            return None, [], None
+        
+        # Use the best match (first result, as they're sorted by relevance)
+        best_match = search_results[0]
+        paper_url = best_match.get('forum_url')
+        
+        logger.debug(f"Found OpenReview match: {best_match.get('title', 'Untitled')}")
+        
+        # Verify the reference against the found paper
+        errors = []
+        
+        # Check title match
+        cited_title = reference.get('title', '').strip()
+        paper_title = best_match.get('title', '').strip()
+        
+        if cited_title and paper_title:
+            similarity = compare_titles_with_latex_cleaning(cited_title, paper_title)
+            if similarity < 0.8:  # Slightly higher threshold for search results
+                from refchecker.utils.error_utils import format_title_mismatch
+                # Clean the cited title for display (remove LaTeX commands like {LLM}s -> LLMs)
+                clean_cited_title = strip_latex_commands(cited_title)
+                details = format_title_mismatch(clean_cited_title, paper_title) + f" (similarity: {similarity:.2f})"
+                errors.append({
+                    "warning_type": "title",
+                    "warning_details": details
+                })
+        
+        # Check authors
+        cited_authors = reference.get('authors', [])
+        paper_authors = best_match.get('authors', [])
+        
+        if cited_authors and paper_authors:
+            # Convert to list format if needed
+            if isinstance(cited_authors, str):
+                cited_authors = [author.strip() for author in cited_authors.split(',')]
+            if isinstance(paper_authors, str):
+                paper_authors = [author.strip() for author in paper_authors.split(',')]
+            
+            # Use the existing author comparison function
+            match, error_msg = compare_authors(cited_authors, paper_authors)
+            if not match and error_msg:
+                errors.append({
+                    "warning_type": "author", 
+                    "warning_details": error_msg
+                })
+        
+        # Check year
+        cited_year = reference.get('year')
+        paper_year = best_match.get('year')
+        
+        if cited_year and paper_year:
+            try:
+                cited_year_int = int(cited_year)
+                paper_year_int = int(paper_year)
+                
+                is_different, year_message = is_year_substantially_different(cited_year_int, paper_year_int)
+                if is_different and year_message:
+                    from refchecker.utils.error_utils import format_year_mismatch
+                    errors.append({
+                        "warning_type": "year",
+                        "warning_details": format_year_mismatch(cited_year_int, paper_year_int)
+                    })
+            except (ValueError, TypeError):
+                pass  # Skip year validation if conversion fails
+        
+        # Check venue if provided in reference
+        cited_venue = reference.get('venue', '').strip()
+        paper_venue = best_match.get('venue', '').strip()
+        
+        if cited_venue and paper_venue:
+            if are_venues_substantially_different(cited_venue, paper_venue):
+                from refchecker.utils.error_utils import format_venue_mismatch
+                errors.append({
+                    "warning_type": "venue",
+                    "warning_details": format_venue_mismatch(cited_venue, paper_venue)
+                })
+        
+        # Create verified data structure
+        verified_data = {
+            'title': best_match.get('title', cited_title),
+            'authors': best_match.get('authors', cited_authors),
+            'year': best_match.get('year', cited_year),
+            'venue': best_match.get('venue', cited_venue),
+            'url': paper_url,
+            'abstract': best_match.get('abstract', ''),
+            'keywords': best_match.get('keywords', []),
+            'openreview_metadata': best_match,
+            'verification_source': 'OpenReview (search)'
+        }
+        
+        logger.debug(f"OpenReview search verification completed for: {paper_url}")
+        return verified_data, errors, paper_url
+    
+    def _is_likely_openreview_venue(self, venue: str) -> bool:
+        """
+        Check if a venue suggests the paper might be on OpenReview
+        
+        Args:
+            venue: Venue string from reference
+            
+        Returns:
+            True if venue suggests OpenReview
+        """
+        if not venue:
+            return False
+        
+        venue_lower = venue.lower()
+        
+        # Common venues that use OpenReview
+        openreview_venues = [
+            'iclr', 'international conference on learning representations',
+            'neurips', 'neural information processing systems', 'nips',
+            'icml', 'international conference on machine learning',
+            'iclr workshop', 'neurips workshop', 'icml workshop',
+            'aaai', 'ijcai', 'aistats'
+        ]
+        
+        for or_venue in openreview_venues:
+            if or_venue in venue_lower:
+                return True
+        
+        return False
+    
+    def search_paper(self, title: str, authors: List[str] = None, year: int = None) -> List[Dict[str, Any]]:
+        """
+        Search for papers on OpenReview by title, authors, and/or year
+        
+        Args:
+            title: Paper title to search for
+            authors: List of author names (optional)
+            year: Publication year (optional)
+            
+        Returns:
+            List of matching paper metadata dictionaries
+        """
+        if not title or not title.strip():
+            return []
+            
+        logger.debug(f"Searching OpenReview for: {title}")
+        
+        # Clean title for search
+        search_title = clean_title_for_search(title)
+        
+        # Try API search first
+        results = self._search_via_api(search_title, authors, year)
+        if results:
+            return results
+        
+        # If API search fails, try web search as fallback
+        return self._search_via_web(search_title, authors, year)
+    
+    def _search_via_api(self, title: str, authors: List[str] = None, year: int = None) -> List[Dict[str, Any]]:
+        """
+        Search using OpenReview API
+        
+        Args:
+            title: Clean title to search for
+            authors: List of author names (optional)
+            year: Publication year (optional)
+            
+        Returns:
+            List of matching paper dictionaries
+        """
+        try:
+            # The OpenReview API requires specific parameters
+            # We'll search by content.title or content.venue (for venue-based search)
+            search_params = {
+                'limit': 20,  # Limit results to avoid overwhelming the API
+                'details': 'directReplies'  # Get basic details
+            }
+            
+            # Try searching by venue first if year suggests recent conferences
+            if year and year >= 2017:  # OpenReview started around 2017
+                venues_by_year = {
+                    2025: ['ICLR 2025'],
+                    2024: ['ICLR 2024', 'NeurIPS 2024', 'ICML 2024'],
+                    2023: ['ICLR 2023', 'NeurIPS 2023', 'ICML 2023'],
+                    2022: ['ICLR 2022', 'NeurIPS 2022', 'ICML 2022'],
+                    2021: ['ICLR 2021', 'NeurIPS 2021', 'ICML 2021'],
+                    2020: ['ICLR 2020', 'NeurIPS 2020', 'ICML 2020'],
+                    2019: ['ICLR 2019', 'NeurIPS 2019', 'ICML 2019'],
+                    2018: ['ICLR 2018', 'NeurIPS 2018', 'ICML 2018'],
+                    2017: ['ICLR 2017']
+                }
+                
+                possible_venues = venues_by_year.get(year, [])
+                
+                results = []
+                for venue in possible_venues:
+                    # Search by venue and then filter by title
+                    venue_params = search_params.copy()
+                    venue_params['content.venue'] = venue
+                    
+                    api_url = f"{self.api_url}/notes"
+                    response = self._respectful_request(api_url, params=venue_params)
+                    
+                    if response and response.status_code == 200:
+                        try:
+                            data = response.json()
+                            if 'notes' in data and data['notes']:
+                                for note in data['notes']:
+                                    try:
+                                        metadata = self._parse_api_response(note)
+                                        if metadata and self._is_good_match(metadata, title, authors, year):
+                                            results.append(metadata)
+                                            if len(results) >= 5:  # Limit results
+                                                break
+                                    except Exception as e:
+                                        logger.debug(f"Error parsing note: {e}")
+                                        continue
+                                
+                                if results:
+                                    break  # Found results, no need to search other venues
+                                    
+                        except (json.JSONDecodeError, KeyError) as e:
+                            logger.debug(f"Failed to parse venue search response: {e}")
+                            continue
+                    else:
+                        logger.debug(f"Venue search failed for {venue}: {response.status_code if response else 'No response'}")
+                
+                if results:
+                    logger.debug(f"OpenReview API search found {len(results)} matches via venue search")
+                    return results
+            
+            # If venue search didn't work, try other approaches
+            # OpenReview API is quite restrictive, so we might need to fall back to web scraping
+            logger.debug("OpenReview API venue search returned no results, trying web search")
+            return []
+            
+        except Exception as e:
+            logger.debug(f"OpenReview API search error: {e}")
+            return []
+    
+    def _search_via_web(self, title: str, authors: List[str] = None, year: int = None) -> List[Dict[str, Any]]:
+        """
+        Search using OpenReview web interface (fallback)
+        
+        Args:
+            title: Clean title to search for
+            authors: List of author names (optional)
+            year: Publication year (optional)
+            
+        Returns:
+            List of matching paper dictionaries
+        """
+        try:
+            # Build search URL
+            search_query = title.replace(' ', '+')
+            search_url = f"{self.base_url}/search?term={search_query}"
+            
+            response = self._respectful_request(search_url)
+            if not response or response.status_code != 200:
+                return []
+            
+            # Parse search results page
+            soup = BeautifulSoup(response.text, 'html.parser')
+            
+            # Look for paper links in search results
+            # OpenReview search results typically contain links to forum pages
+            results = []
+            
+            # Find links that look like OpenReview paper URLs
+            for link in soup.find_all('a', href=True):
+                href = link.get('href', '')
+                if '/forum?id=' in href:
+                    paper_id = self.extract_paper_id(href)
+                    if paper_id:
+                        # Get full metadata for this paper
+                        metadata = self.get_paper_metadata(paper_id)
+                        if metadata and self._is_good_match(metadata, title, authors, year):
+                            results.append(metadata)
+                            if len(results) >= 5:  # Limit results
+                                break
+            
+            logger.debug(f"OpenReview web search found {len(results)} matches")
+            return results
+            
+        except Exception as e:
+            logger.debug(f"OpenReview web search error: {e}")
+            return []
+    
+    def _is_good_match(self, metadata: Dict[str, Any], search_title: str, authors: List[str] = None, year: int = None) -> bool:
+        """
+        Check if the found paper is a good match for the search criteria
+        
+        Args:
+            metadata: Paper metadata from OpenReview
+            search_title: Title we're searching for
+            authors: Authors we're looking for (optional)
+            year: Year we're looking for (optional)
+            
+        Returns:
+            True if it's a good match
+        """
+        paper_title = metadata.get('title', '')
+        if not paper_title:
+            return False
+        
+        # Check title similarity
+        title_similarity = calculate_title_similarity(search_title, paper_title)
+        if title_similarity < 0.7:  # Require at least 70% similarity
+            return False
+        
+        # Check year if provided
+        if year:
+            paper_year = metadata.get('year')
+            if paper_year and abs(int(paper_year) - year) > 1:  # Allow 1 year difference
+                return False
+        
+        # Check authors if provided
+        if authors and len(authors) > 0:
+            paper_authors = metadata.get('authors', [])
+            if paper_authors:
+                # Check if at least one author matches
+                author_match = False
+                for search_author in authors[:2]:  # Check first 2 authors
+                    for paper_author in paper_authors[:3]:  # Check first 3 paper authors
+                        if is_name_match(search_author, paper_author):
+                            author_match = True
+                            break
+                    if author_match:
+                        break
+                
+                if not author_match:
+                    return False
+        
+        return True
+
+    def search_by_title(self, title: str, max_results: int = 5) -> List[Dict[str, Any]]:
+        """
+        Search OpenReview for papers by title using the working search API.
+        
+        Args:
+            title: Paper title to search for
+            max_results: Maximum number of results to return
+            
+        Returns:
+            List of paper data dictionaries
+        """
+        try:
+            # Use OpenReview's search API with term parameter (this works!)
+            params = {
+                'term': title,
+                'limit': max_results
+            }
+            
+            response = self._respectful_request(f"{self.api_url}/notes/search", params=params)
+            if not response or response.status_code != 200:
+                logger.debug(f"OpenReview search API failed with status {response.status_code if response else 'None'}")
+                return []
+            
+            data = response.json()
+            papers = []
+            
+            for note in data.get('notes', []):
+                # Filter to exact or close title matches
+                note_title = note.get('content', {}).get('title', '')
+                if self._is_title_match(title, note_title):
+                    paper_data = self._parse_api_response(note)
+                    if paper_data:
+                        papers.append(paper_data)
+            
+            logger.debug(f"OpenReview search found {len(papers)} matching papers for '{title}'")
+            return papers
+            
+        except Exception as e:
+            logger.error(f"Error searching OpenReview by title '{title}': {e}")
+            return []
+
+    def _is_title_match(self, search_title: str, found_title: str, threshold: float = 0.8) -> bool:
+        """
+        Check if two titles match closely enough.
+        
+        Args:
+            search_title: Title we're searching for
+            found_title: Title found in search results
+            threshold: Similarity threshold (0.0 to 1.0)
+            
+        Returns:
+            True if titles match closely enough
+        """
+        if not search_title or not found_title:
+            return False
+        
+        # Exact match
+        if search_title.lower().strip() == found_title.lower().strip():
+            return True
+        
+        # Check if one contains the other (for cases where one is longer)
+        search_clean = search_title.lower().strip()
+        found_clean = found_title.lower().strip()
+        
+        if search_clean in found_clean or found_clean in search_clean:
+            return True
+        
+        # Use similarity calculation from text_utils
+        try:
+            from refchecker.utils.text_utils import calculate_title_similarity
+            similarity = calculate_title_similarity(search_title, found_title)
+            return similarity >= threshold
+        except ImportError:
+            # Fallback to simple word matching
+            search_words = set(search_clean.split())
+            found_words = set(found_clean.split())
+            
+            if not search_words or not found_words:
+                return False
+            
+            intersection = search_words.intersection(found_words)
+            union = search_words.union(found_words)
+            
+            jaccard_similarity = len(intersection) / len(union) if union else 0
+            return jaccard_similarity >= threshold
+
+    def verify_reference_by_search(self, reference: Dict[str, Any]) -> Tuple[Optional[Dict[str, Any]], List[Dict[str, Any]], Optional[str]]:
+        """
+        Verify a reference by searching OpenReview (for papers without URLs).
+        
+        Args:
+            reference: Reference data dictionary
+            
+        Returns:
+            Tuple of (verified_data, errors_and_warnings, debug_info)
+        """
+        title = reference.get('title', '').strip()
+        if not title:
+            return None, [], "No title provided for search"
+        
+        # Search for the paper
+        search_results = self.search_by_title(title)
+        
+        if not search_results:
+            return None, [], f"No papers found on OpenReview for title: {title}"
+        
+        # Take the best match (first result, as search is already filtered)
+        best_match = search_results[0]
+        
+        # Use the existing verify_reference method with the found URL
+        forum_url = best_match.get('forum_url')
+        if forum_url:
+            # Create a reference with the OpenReview URL for verification
+            reference_with_url = reference.copy()
+            reference_with_url['url'] = forum_url
+            
+            return self.verify_reference(reference_with_url)
+        
+        # If no URL, return the metadata as verification
+        return best_match, [], f"Found on OpenReview: {best_match.get('title')}"