Gazetteer-Augmented NER Model for Rare Entity Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional natural language processing (NLP) systems, particularly named entity recognition (NER) systems, face challenges in accurately identifying rare or unpopular named entities due to limited training data, leading to poor performance in recognizing new or uncommon entities such as song titles or brand names, which are not well-represented in the general population's usage.
Innovation Solution
The NLP system incorporates a gazetteer-augmented NER model that uses a catalog of tokens and word groupings to enhance the recognition of named entities by providing a probability boost based on gazetteer data, combined with word and character embeddings, and adjusts the gazetteer output weights during training to improve recall and precision without compromising computation speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional NER systems are trained on general population data, then computation speed is maintained, but recognition accuracy of rare or unpopular named entities deteriorates
Solution Approach 1:
The patent combines conventional NER models trained on general population data with gazetteer-based features containing curated lists of rare and unpopular named entities. This merging allows the system to leverage both the computational efficiency of general models and the specialized knowledge of gazetteers, improving recognition accuracy for rare entities without sacrificing overall computation speed.
Solution Approach 2:
Gazetteers serve as an intermediary data source that bridges the gap between general training data and specific rare named entities. The gazetteer features act as a mediator, providing targeted information about uncommon entities to the NER system without requiring extensive customization of the entire model architecture.
2Measurement precision
If customized NER systems are created for rare entities, then recognition precision improves, but system complexity and development costs increase
Solution Approach 1:
The patent implements a universal NER system that can handle both common and rare named entities through a single architecture. By integrating gazetteer features into the conventional NER framework, the system achieves multi-functionality, handling diverse entity types without requiring separate customized models for rare entities.
Solution Approach 2:
The system modifies the feature parameters of the conventional NER model by incorporating gazetteer-based features. This parameter change allows the model to adapt to rare entities by adjusting its input features rather than changing the entire model structure, reducing complexity while improving precision.
3Reliability
If gazetteer data is integrated into NER model, then recall of rare named entities improves, but computation time may increase
Solution Approach 1:
Gazetteers are pre-computed and curated before being integrated into the NER system. This preliminary action allows the system to quickly lookup rare entity information during inference without performing computationally intensive processing, improving recall while minimizing additional computation time.
Solution Approach 2:
The patent replaces the need for extensive model retraining and customization with a more efficient mechanism: integrating pre-computed gazetteer features. This substitution reduces the computational burden during both training and inference, maintaining fast processing speeds while improving entity recall.
Data Source
AI summary
Features are disclosed for training and using named entity recognition models based on gazetteer information. A named entity recognition model can be trained with a gazetteer output at a layer of the model to provide deterministic data in the probabilistic model. The named entity recognition model can recognize named entities based on the word embedding and the gazetteer output. The named entity recognition model can tune the gazetteer output to include false positive name entities such that the gazetteer output is not deterministic of the output of the model. In some embodiments, the named entity recognition model can be tuned so as to adjust the gazetteer output.