Named Entity Recognition Model Positional Tag Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional named entity recognition (NER) models struggle to accurately distinguish between different positional tags, such as 'B-Name' and 'I-Name', leading to suboptimal recognition performance due to the lack of positional information in the training process.
Innovation Solution
The method involves using a pre-trained language model and an attention mechanism with a weight matrix that incorporates positional information for tokens corresponding to tags, allowing the model to better understand the positions of tokens within named entities, thereby improving recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional NER models use sequence tagging with pre-trained language models, then the model can perform basic named entity recognition, but the model cannot distinguish between different positional tags (such as B-Name and I-_name) because positional information is not provided in the training process
Solution Approach 1:
The patent applies preliminary action by pre-marking training texts with positional information before training the model. Positional indication tokens are inserted into the training data to indicate the position of tokens within named entities, allowing the model to learn positional distinctions during training. This preliminary preparation of training data enables the model to accurately distinguish between B-Name and I-_name tags without requiring complex post-processing.
Solution Approach 2:
The patent introduces positional indication tokens as intermediaries between the token sequence and the tag annotations. These special tokens serve as mediators that carry positional information about tokens within named entities, allowing the attention mechanism to properly distinguish between different positional tags. The positional indication tokens act as a bridge that transmits positional context to the model during training.
2Ease of manufacture
If the model outputs tags as numbers without semantic meaning, then the training process is simplified, but the model cannot understand what the tags represent and recognition performance is limited
Solution Approach 1:
The patent introduces tag annotation information as an intermediary layer between the numerical tag outputs and the actual named entity types. The tag annotation includes semantic descriptions and positional indications that explain what each tag represents. This intermediary layer allows the model to maintain simple numerical outputs during training while simultaneously understanding the semantic meaning of tags through the accompanying annotations.
Solution Approach 2:
The patent implements feedback by providing tag annotation information that feeds back into the training process. The tag annotations, which include semantic meanings and positional indications, are used to compute loss and update model parameters. This feedback mechanism ensures that the model learns not only to output tags but also to understand their semantic meaning and positional context, thereby improving recognition performance.
Data Source
AI summary
A method and an apparatus are provided for training a named entity recognition (NER) model. By constructing tag annotations for tags and causing the tag annotations to contain information for indicating the positions of tokens in named entities, corresponding to the tags, respectively, in the process of training the NER model, the NER model can better understand the different positions of different tokens in the same named entity, so that the trained NER model can more accurately recognize named entities.