Current research focuses on applying ML/NLP enhancements to improve case analysis capabilities. Specfically evaluating semantic similarity, named entity recognition, and content sanitization to complement existing pattern-based extraction, clustering, and visualizations. Additionally, ML models will be run on cases within each extraction feature to identify similarities, differences, and distinguishing patterns between cases in each stat category (e.g. FBI cases vs Local Law Enforcement investigations, Stranger cases vs Family or Teacher cases, possible offender patterns across ages)
The core system demonstrates pattern-based extraction works well for structured case data. This phase explores where ML can add value: understanding semantic relationships, extracting entities regex might miss, generating case summaries that preserve insights while reducing exposure to harmful content, and analyzing case groups to identify distinguishing features, similarities, and differences between groups.
sentence-transformers — Semantic embeddings (all-MiniLM-L6-v2)spaCy — Named entity recognition (en_core_web_sm)transformers — Summarization models (BART/T5)ML components are optional enhancements, not replacements. Pattern-based extraction remains primary; ML provides additional capabilities. System reverts gracefully if ML models are unavailable or undesired.
Processing flow: Case Text → Pattern extraction → ML enhancement (optional) → Storage. Components are modular and can be evaluated independently.
Analysis Execution: All ML analysis (comparative case group analysis, semantic similarity, clustering) is pre-computed offline, not performed on live CaseLinker instances. Results are stored and served via API endpoints, ensuring performance and consistency without impacting real-time case processing.
Deployment: ML dependencies are available in requirements-ml.txt
for local development. Production builds use requirements.txt (ML deps commented)
for faster deployment; all ML code remains in the repository.