Entity Recognition Model Training via Vector Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current named entity recognition models, particularly BiLSTM-CRF, face challenges in achieving high accuracy for semantic understanding in natural language processing, leading to incomplete user expression in dialogue systems.
Innovation Solution
An entity recognition model training method that involves obtaining feature vectors from training text, merging one-hot vectors, and adding them to word vectors to improve model accuracy, along with probability calculations and transition probability adjustments to refine recognition results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If BiLSTM-CRF model is used for named entity recognition, then the model becomes more common and widely applicable, but the accuracy in semantic understanding remains ideal and information accuracy for word slots is low
Solution Approach 1:
The patent merges one-hot vectors from database matching with word vectors to create enhanced feature representations. This combination allows the model to leverage both lexical information (from word vectors) and semantic/ contextual information (from one-hot vectors representing entities, phrases, or concepts), thereby improving semantic understanding accuracy while maintaining model applicability
Solution Approach 2:
The patent introduces an intermediary feature vector space that combines one-hot vectors and word vectors. This intermediary representation serves as a bridge between raw input text and final entity recognition outputs, enabling the model to capture nuanced semantic relationships that neither vector type alone could provide, thus resolving the accuracy limitation
2Measurement precision
If feature vectors are merged with one-hot vectors and added to word vectors, then entity recognition accuracy is improved, but model complexity increases
Solution Approach 1:
The patent performs preliminary feature extraction and vector merging during the training phase, where feature vectors are pre-computed and combined with word vectors to create enriched representations. This preliminary action allows the complex merging operation to be performed once during training, after which the model can use these pre-computed features during inference without repeatedly performing complex calculations, thus managing model complexity while maintaining high accuracy
Data Source
AI summary
The present disclosure discloses an entity recognition model training method and an entity recognition method as well as an apparatus using them. The entity recognition model training method includes: obtaining a training text and matching the training text with a database to obtain a plurality of matching results; processing the matching results to obtain a plurality of feature vectors corresponding to the matching results; obtaining a word vector of each word in the training text by processing the training text; and training an initial entity recognition model based on the feature vector and the word vector to obtain an entity recognition model. By using this training manner, the entity recognition model obtained can have an improved accuracy of entity recognition.