| | @@ -0,0 +1,193 @@ |
| 1 | +"""Taxonomy classifier for planning entity extraction.
|
| 2 | +
|
| 3 | +Bridges raw knowledge graph entities (person, technology, concept) into
|
| 4 | +planning-ready structures (goals, requirements, decisions, risks).
|
| 5 | +"""
|
| 6 | +
|
| 7 | +import logging
|
| 8 | +from typing import Any, Dict, List, Optional
|
| 9 | +
|
| 10 | +from video_processor.models import PlanningEntity, PlanningEntityType
|
| 11 | +
|
| 12 | +logger = logging.getLogger(__name__)
|
| 13 | +
|
| 14 | +# Keyword rules for heuristic classification. Each tuple is
|
| 15 | +# (PlanningEntityType, list-of-keywords). Order matters — first match wins.
|
| 16 | +_KEYWORD_RULES: List[tuple] = [
|
| 17 | + (PlanningEntityType.GOAL, ["goal", "objective", "aim", "target outcome"]),
|
| 18 | + (
|
| 19 | + PlanningEntityType.REQUIREMENT,
|
| 20 | + ["must", "should", "requirement", "need", "required"],
|
| 21 | + ),
|
| 22 | + (
|
| 23 | + PlanningEntityType.CONSTRAINT,
|
| 24 | + ["constraint", "limitation", "restrict", "cannot", "must not"],
|
| 25 | + ),
|
| 26 | + (
|
| 27 | + PlanningEntityType.DECISION,
|
| 28 | + ["decided", "decision", "chose", "selected", "agreed"],
|
| 29 | + ),
|
| 30 | + (PlanningEntityType.RISK, ["risk", "concern", "worry", "danger", "threat"]),
|
| 31 | + (
|
| 32 | + PlanningEntityType.ASSUMPTION,
|
| 33 | + ["assume", "assumption", "expecting", "presume"],
|
| 34 | + ),
|
| 35 | + (
|
| 36 | + PlanningEntityType.DEPENDENCY,
|
| 37 | + ["depends", "dependency", "relies on", "prerequisite", "blocked"],
|
| 38 | + ),
|
| 39 | + (
|
| 40 | + PlanningEntityType.MILESTONE,
|
| 41 | + ["milestone", "deadline", "deliverable", "release", "launch"],
|
| 42 | + ),
|
| 43 | + (
|
| 44 | + PlanningEntityType.TASK,
|
| 45 | + ["task", "todo", "action item", "work item", "implement"],
|
| 46 | + ),
|
| 47 | + (PlanningEntityType.FEATURE, ["feature", "capability", "functionality"]),
|
| 48 | +]
|
| 49 | +
|
| 50 | +
|
| 51 | +class TaxonomyClassifier:
|
| 52 | + """Classifies raw knowledge graph entities into planning taxonomy types."""
|
| 53 | +
|
| 54 | + def __init__(self, provider_manager: Optional[Any] = None):
|
| 55 | + self.pm = provider_manager
|
| 56 | +
|
| 57 | + # ------------------------------------------------------------------
|
| 58 | + # Public API
|
| 59 | + # ------------------------------------------------------------------
|
| 60 | +
|
| 61 | + def classify_entities(
|
| 62 | + self,
|
| 63 | + entities: List[Dict],
|
| 64 | + relationships: List[Dict],
|
| 65 | + ) -> List[PlanningEntity]:
|
| 66 | + """Classify extracted entities into planning entity types.
|
| 67 | +
|
| 68 | + Uses heuristic classification first, then LLM refinement if a
|
| 69 | + provider manager is available.
|
| 70 | + """
|
| 71 | + planning_entities: List[PlanningEntity] = []
|
| 72 | +
|
| 73 | + # Step 1: heuristic classification
|
| 74 | + for entity in entities:
|
| 75 | + planning_type = self._heuristic_classify(entity, relationships)
|
| 76 | + if planning_type:
|
| 77 | + descs = entity.get("descriptions", [])
|
| 78 | + planning_entities.append(
|
| 79 | + PlanningEntity(
|
| 80 | + name=entity["name"],
|
| 81 | + planning_type=planning_type,
|
| 82 | + description="; ".join(descs[:2]),
|
| 83 | + source_entities=[entity["name"]],
|
| 84 | + )
|
| 85 | + )
|
| 86 | +
|
| 87 | + # Step 2: LLM refinement (if provider available)
|
| 88 | + if self.pm and entities:
|
| 89 | + llm_classified = self._llm_classify(entities, relationships)
|
| 90 | + planning_entities = self._merge_classifications(planning_entities, llm_classified)
|
| 91 | +
|
| 92 | + return planning_entities
|
| 93 | +
|
| 94 | + def organize_by_workstream(
|
| 95 | + self, planning_entities: List[PlanningEntity]
|
| 96 | + ) -> Dict[str, List[PlanningEntity]]:
|
| 97 | + """Group planning entities into logical workstreams by type."""
|
| 98 | + workstreams: Dict[str, List[PlanningEntity]] = {}
|
| 99 | + for pe in planning_entities:
|
| 100 | + group = pe.planning_type.value + "s"
|
| 101 | + workstreams.setdefault(group, []).append(pe)
|
| 102 | + return workstreams
|
| 103 | +
|
| 104 | + # ------------------------------------------------------------------
|
| 105 | + # Heuristic classification
|
| 106 | + # ------------------------------------------------------------------
|
| 107 | +
|
| 108 | + def _heuristic_classify(
|
| 109 | + self,
|
| 110 | + entity: Dict,
|
| 111 | + relationships: List[Dict], # noqa: ARG002 — reserved for future rules
|
| 112 | + ) -> Optional[PlanningEntityType]:
|
| 113 | + """Rule-based classification from entity type and description keywords."""
|
| 114 | + desc_lower = " ".join(entity.get("descriptions", [])).lower()
|
| 115 | +
|
| 116 | + for planning_type, keywords in _KEYWORD_RULES:
|
| 117 | + if any(kw in desc_lower for kw in keywords):
|
| 118 | + return planning_type
|
| 119 | +
|
| 120 | + return None
|
| 121 | +
|
| 122 | + # ------------------------------------------------------------------
|
| 123 | + # LLM classification
|
| 124 | + # ------------------------------------------------------------------
|
| 125 | +
|
| 126 | + def _llm_classify(
|
| 127 | + self, entities: List[Dict], relationships: List[Dict]
|
| 128 | + ) -> List[PlanningEntity]:
|
| 129 | + """Use LLM to classify entities into planning types."""
|
| 130 | + entity_summaries = []
|
| 131 | + for e in entities[:50]: # limit to avoid token overflow
|
| 132 | + descs = e.get("descriptions", [])
|
| 133 | + desc_str = "; ".join(descs[:2]) if descs else "no description"
|
| 134 | + entity_summaries.append(f"- {e['name']} ({e.get('type', 'concept')}): {desc_str}")
|
| 135 | +
|
| 136 | + prompt = (
|
| 137 | + "Classify these entities from a knowledge graph into planning categories.\n\n"
|
| 138 | + "Entities:\n" + "\n".join(entity_summaries) + "\n\n"
|
| 139 | + "Categories: goal, requirement, constraint, decision, risk, assumption, "
|
| 140 | + "dependency, milestone, task, feature\n\n"
|
| 141 | + "For each entity that fits a planning category, return JSON:\n"
|
| 142 | + '[{"name": "...", "planning_type": "...", "priority": "high|medium|low"}]\n\n'
|
| 143 | + "Only include entities that clearly fit a planning category. "
|
| 144 | + "Skip entities that are just people, technologies, or general concepts. "
|
| 145 | + "Return ONLY the JSON array."
|
| 146 | + )
|
| 147 | +
|
| 148 | + try:
|
| 149 | + raw = self.pm.chat(
|
| 150 | + [{"role": "user", "content": prompt}],
|
| 151 | + max_tokens=2048,
|
| 152 | + temperature=0.2,
|
| 153 | + )
|
| 154 | + except Exception:
|
| 155 | + logger.warning("LLM classification failed, using heuristic only")
|
| 156 | + return []
|
| 157 | +
|
| 158 | + from video_processor.utils.json_parsing import parse_json_from_response
|
| 159 | +
|
| 160 | + parsed = parse_json_from_response(raw)
|
| 161 | +
|
| 162 | + results: List[PlanningEntity] = []
|
| 163 | + if isinstance(parsed, list):
|
| 164 | + for item in parsed:
|
| 165 | + if isinstance(item, dict) and "name" in item and "planning_type" in item:
|
| 166 | + try:
|
| 167 | + ptype = PlanningEntityType(item["planning_type"])
|
| 168 | + results.append(
|
| 169 | + PlanningEntity(
|
| 170 | + name=item["name"],
|
| 171 | + planning_type=ptype,
|
| 172 | + priority=item.get("priority"),
|
| 173 | + source_entities=[item["name"]],
|
| 174 | + )
|
| 175 | + )
|
| 176 | + except ValueError:
|
| 177 | + pass
|
| 178 | + return results
|
| 179 | +
|
| 180 | + # ------------------------------------------------------------------
|
| 181 | + # Merge
|
| 182 | + # ------------------------------------------------------------------
|
| 183 | +
|
| 184 | + @staticmethod
|
| 185 | + def _merge_classifications(
|
| 186 | + heuristic: List[PlanningEntity],
|
| 187 | + llm: List[PlanningEntity],
|
| 188 | + ) -> List[PlanningEntity]:
|
| 189 | + """Merge heuristic and LLM classifications. LLM wins on conflicts."""
|
| 190 | + by_name = {pe.name.lower(): pe for pe in heuristic}
|
| 191 | + for pe in llm:
|
| 192 | + by_name[pe.name.lower()] = pe # LLM overrides
|
| 193 | + return list(by_name.values())
|