BERT and GCN Entity Disambiguation for Knowledge Map Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing entity disambiguation technologies face challenges in accurately identifying the correct meaning of an entity in context and associating it with the appropriate concept in a knowledge map, due to the complexity of natural language.
Innovation Solution
A new method using a classification model that encodes entities with a Bidirectional Encoder Representations and Transformers (BERT) model, and utilizes character similarity and semantic similarity as auxiliary features to improve the accuracy of entity disambiguation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Word2Vec similarity feature is used in the classification model, then the model can capture some semantic information, but the semantic information is insufficient and it is difficult to correctly determine semantic level similarity between entity and concept
Solution Approach 1:
The patent changes the parameter of semantic representation from Word2Vec vectors to BERT-generated contextualized vectors. This transformation enables the model to capture richer semantic information and contextual nuances, thereby improving the accuracy of semantic similarity determination between entities and concepts while preserving complete semantic information.
Solution Approach 2:
The patent introduces BERT as an intermediary component between the input text and the classification model. BERT processes the entity and concept texts to generate high-quality semantic representations, acting as a bridge that transforms raw text into meaningful vector representations that preserve complete semantic information for accurate similarity calculation.
2Ease of manufacture
If simple rule is used in candidate concept generation step, then the process is simple, but proper candidate concepts cannot be filtered out resulting in cascading error in subsequent ranking step
Solution Approach 1:
The patent applies preliminary action by using BERT to generate accurate candidate concepts and calculate their semantic similarities before the final classification step. This preliminary processing ensures that high-quality candidate concepts are identified and ranked correctly, preventing cascading errors in subsequent steps while maintaining systematic efficiency.
3Measurement precision
If entity is encoded by BERT model with character similarity and semantic similarity as auxiliary features, then the probability of correctly associating entity with concept increases, but the model complexity increases
Solution Approach 1:
The patent applies universality by using the BERT model to perform multiple functions simultaneously: encoding the entity, encoding the concepts, generating contextualized representations, and providing semantic similarity measurements. This multi-functionality approach improves association accuracy while avoiding the need for separate dedicated components for each function, thereby managing model complexity efficiently.
Data Source
AI summary
A method of presenting prompt information by utilizing a neural network which includes a BERT model and a graph convolutional neural network (GCN), comprising: generating a first vector based on a combination of an entity, a context of the entity, a type of the entity and a part of speech of the context by using BERT model; generating a second vector based on each of predefined concepts by using BERT model; generating a third vector based on a graph which is generated based on the concepts and relationships thereamong, by using GCN; generating a fourth vector by concatenating the second and third vectors; calculating semantic similarity between the entity and each concept based on the first and fourth vectors; determining, based on the first vector and the semantic similarity, that the entity corresponds to one of the concepts; and generating the prompt information based on the determined concept.