luopeixiang/named_entity_recognition
Named Entity Recognition
Named entity recognition is a natural language processing library that implements statistical and neural sequence labeling models to extract entities from text. The toolkit provides implementations for hidden Markov models, conditional random fields, and bidirectional recurrent neural networks combined with conditional random field layers.
The library supports training machine learning models on annotated training corpora using maximum likelihood estimation for parameter and transition structure estimation. It includes ensemble majority voting consensus strategies to combine independent outputs from multiple underlying sequence models, producing reliable consensus predictions.
Features
- Named Entity Recognition - Provides a comprehensive named entity recognition library implementing HMM, CRF, and neural sequence labeling models.
- Bidirectional LSTM Models - Implements a deep learning architecture combining recurrent neural networks to process text sequences in both directions.
- Entity and Relation Extraction - Combines bidirectional neural feature extraction with a conditional random field layer to model both text features and label dependencies.
- Machine Learning Training - Estimates parameters and structures for hidden Markov models, conditional random fields, and recurrent neural networks using labeled training data.
- Hidden Markov Models - Identifies text entities using probabilistic state transitions and observation probabilities estimated from training data via maximum likelihood.
- Natural Language Entity Extraction - Learns non-linear transformations through bidirectional recurrent layers to predict token labels directly from input text sequences.
- Conditional Random Fields - Implements conditional random field models to capture label transition dependencies for sequence tagging tasks.
- Hybrid Neural Random Fields - Combines bidirectional neural sequence feature extraction with a conditional random field layer to model text features and label dependencies.
- Sequence Labeling - Applies statistical sequence labeling models to assign token-level labels and extract entities from text.
- Maximum Likelihood Estimators - Estimates model parameters and transition structures for statistical and neural sequence taggers using maximum likelihood estimation.
- Bidirectional Recurrent Neural Networks - Processes text sequences in both forward and backward directions through neural layers to learn context representations.
- Majority-Vote Ensembles - Combines independent outputs from multiple underlying sequence models using voting strategies to produce reliable consensus predictions.
- Ensemble Prediction Combinations - Combines outputs from multiple underlying models using majority voting strategies to produce reliable consensus predictions.
