Spaces:
Sleeping
Sleeping
| """ | |
| SWOT text parsing service. | |
| Extracts structured SWOT data from markdown/text reports. | |
| Contract: within the four SWOT sections, only lines carrying an [M##] | |
| citation are treated as SWOT items. Uncited bullets (typically action | |
| recommendations the LLM appends) are diverted to `recommendations`, and any | |
| "Data Quality Notes"-style section goes to `data_quality_notes` — neither | |
| leaks into the quadrant tables. | |
| """ | |
| import re | |
| # A SWOT item per the analyzer's required format: starts with [M##] | |
| ITEM_PATTERN = re.compile(r'^\[?M\d{2}\]?', re.IGNORECASE) | |
| def parse_swot_text(text: str) -> dict: | |
| """ | |
| Parse SWOT text into structured sections. | |
| Args: | |
| text: Raw SWOT analysis text with sections marked by headers | |
| Returns: | |
| Dictionary with keys: strengths, weaknesses, opportunities, threats, | |
| recommendations, data_quality_notes. Each contains a list of strings. | |
| """ | |
| sections = { | |
| "strengths": [], | |
| "weaknesses": [], | |
| "opportunities": [], | |
| "threats": [], | |
| "recommendations": [], | |
| "data_quality_notes": [], | |
| } | |
| current_section = None | |
| lines = text.split('\n') | |
| # Regex to match various bullet formats: -, *, •, numbered lists (1., 2.), etc. | |
| bullet_pattern = re.compile(r'^[\s]*[-*•]\s*(.+)$|^[\s]*\d+[.)]\s*(.+)$') | |
| swot_headers = [ | |
| ('strength', 'strengths'), | |
| ('weakness', 'weaknesses'), | |
| ('opportunit', 'opportunities'), | |
| ('threat', 'threats'), | |
| ('recommendation', 'recommendations'), | |
| ('data quality', 'data_quality_notes'), | |
| ] | |
| for line in lines: | |
| line = line.strip() | |
| if not line: | |
| continue | |
| lower_line = line.lower() | |
| # Check for section headers (with various formats: ##, **, :, etc.) | |
| clean_lower = re.sub(r'[#*_:\[\]()]', '', lower_line).strip() | |
| header_matched = False | |
| for keyword, section_key in swot_headers: | |
| if keyword in clean_lower and len(clean_lower) < 50: | |
| current_section = section_key | |
| after_header = _extract_after_header(line, keyword) | |
| if after_header: | |
| _route_item(sections, current_section, after_header) | |
| header_matched = True | |
| break | |
| if header_matched: | |
| continue | |
| # Unrecognized markdown header: never emit as content | |
| if line.startswith('#'): | |
| current_section = None | |
| continue | |
| # If we're in a section, try to extract content | |
| if current_section: | |
| match = bullet_pattern.match(line) | |
| if match: | |
| item = match.group(1) or match.group(2) | |
| if item and item.strip(): | |
| _route_item(sections, current_section, item.strip()) | |
| elif not _is_header_line(line) and len(line) > 10: | |
| _route_item(sections, current_section, line) | |
| return sections | |
| def _route_item(sections: dict, current_section: str, item: str): | |
| """ | |
| Place an item per the edge contract: SWOT quadrants only accept | |
| [M##]-cited lines; uncited lines in a quadrant are recommendations. | |
| """ | |
| if current_section in ("recommendations", "data_quality_notes"): | |
| sections[current_section].append(item) | |
| elif ITEM_PATTERN.match(item): | |
| sections[current_section].append(item) | |
| else: | |
| sections["recommendations"].append(item) | |
| def _extract_after_header(line: str, keyword: str) -> str: | |
| """Extract content that appears after a section header on the same line.""" | |
| lower = line.lower() | |
| idx = lower.find(keyword) | |
| if idx == -1: | |
| return "" | |
| end_idx = idx + len(keyword) | |
| while end_idx < len(line) and line[end_idx].isalpha(): | |
| end_idx += 1 | |
| remainder = line[end_idx:].strip() | |
| remainder = re.sub(r'^[:\-–—\s]+', '', remainder).strip() | |
| remainder = re.sub(r'^[#*_]+\s*', '', remainder).strip() | |
| if len(remainder) > 10 and not remainder.lower().startswith(('strength', 'weakness', 'opportunit', 'threat')): | |
| return remainder | |
| return "" | |
| def _is_header_line(line: str) -> bool: | |
| """Check if a line appears to be a header rather than content.""" | |
| clean = re.sub(r'[#*_:\-–—\[\]()]', '', line).strip() | |
| if len(clean) < 5: | |
| return True | |
| if line.rstrip().endswith(':'): | |
| return True | |
| return False | |