Symbolic AI Framework Extracts Incident Facts from Police Reports
A research paper on arXiv proposes a symbolic AI framework to convert unstructured narratives in law enforcement reports into evidence-linked facts. The method uses redaction, semantic parsing, predicate mapping to an ontology, and reasoning to build temporal graphs from text. Evaluated on 450 property crime reports, 54.1% of extracted events had a confidence score of at least 0.80, and 93.7% were mapped through the PropBank–VerbNet–WordNet semantic path. Human review achieved 100% agreement on incident initiation, stolen items, and temporal cues. The framework aims to reduce manual reading for review, training, and investigations.
Key facts
- Framework uses symbolic methods for converting narratives into evidence-linked facts.
- Evaluated on 450 property crime reports.
- 54.1% of extracted events had confidence ≥ 0.80.
- 93.7% mapped through PropBank–VerbNet–WordNet semantic path.
- 100% agreement on incident initiation, stolen items, and temporal cues.
- Redacts personal identifiers before processing.
- Builds temporal graphs with time cues and domain axioms.
- Published on arXiv with ID 2605.15978.
Entities
Institutions
- arXiv
- PropBank
- VerbNet
- WordNet