Struct NerRecognizer 

pub struct NerRecognizer { /* private fields */ }

NER-based recognizer using ONNX Runtime

Status: ✅ Fully operational with complete ONNX Runtime integration

This recognizer uses transformer-based Named Entity Recognition models for contextual PII detection. It automatically loads and runs ONNX models with:

• Tokenization with HuggingFace tokenizers
• ONNX Runtime inference with optimizations
• BIO tag parsing for entity span extraction
• Thread-safe session management

To enable NER:

1. Export your NER model to ONNX format using scripts/export_ner_model.py
2. Set model_path to point to your .onnx file
3. Optionally provide tokenizer_path or place tokenizer.json in the same directory

Without a model, this recognizer gracefully returns empty results and the system falls back to pattern-based detection (36+ entity types).

Implementations

impl NerRecognizer

pub fn from_file<P: AsRef<Path>>(model_path: P) -> Result<Self>

Create a new NER recognizer from a model file

pub fn from_config(config: NerConfig) -> Result<Self>

Create a new NER recognizer from configuration

pub fn config(&self) -> &NerConfig

Get the configuration

pub fn is_available(&self) -> bool

Check if NER is available (model and tokenizer loaded)

Trait Implementations

impl Debug for NerRecognizer

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Recognizer for NerRecognizer

fn name(&self) -> &str

Get the name of this recognizer

fn supported_entities(&self) -> &[EntityType]

Get the entity types this recognizer can detect

fn analyze(&self, text: &str, _language: &str) -> Result<Vec<RecognizerResult>>

Analyze text and return detected entities

fn supports_language(&self, language: &str) -> bool

Check if this recognizer supports the given language

fn min_score(&self) -> f32

Get the minimum confidence score for this recognizer