Machine Learning Named Entity Extraction via Semantic Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing number of named entities, such as chemical substance names and company/personal names, with varying representations poses a challenge for effective extraction from text data using traditional dictionary-based methods, as they often have different names for the same entity, making it difficult to define and extract them accurately.
Innovation Solution
A learning method involving a two-phase approach where a machine learning model learns to extract named entities through a first phase focusing on encoding and a second phase that simultaneously learns both encoding and decoding to handle paraphrased entities, using techniques like word embeddings and long short-term memory networks to generate distributed representations and probability distributions for accurate extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional dictionary-based methods are used for named entity extraction, then the extraction process is simple and fast, but the accuracy decreases when dealing with named entities having different representations or paraphrases
Solution Approach 1:
The patent transforms the extraction approach from exact string matching to semantic similarity comparison by changing the representation parameters. It uses distributed representations (word embeddings) to convert named entities into vector spaces where semantic similarity can be measured, allowing entities with different representations to be recognized as the same entity based on their semantic proximity in the vector space.
Solution Approach 2:
The patent replaces the mechanical dictionary-matching system with a machine learning-based semantic comparison system. Instead of relying on pre-defined dictionary entries and exact string matching, it employs trained models (such as BERT or other contextual embedding models) to compute semantic similarities dynamically, enabling accurate recognition of paraphrased entities.
2Measurement precision
If machine learning models are trained to handle paraphrased entities, then the entity recognition accuracy improves, but the device complexity and training time increase
Solution Approach 1:
The patent creates a universal extraction framework that can handle multiple types of named entities (persons, organizations, locations, chemical substances, etc.) and various representation forms (exact matches, paraphrases, synonyms, different nomenclatures) using a single machine learning model. The model is designed to be entity-type-agnostic, focusing on semantic similarity rather than entity category, which simplifies the overall system architecture despite the increased accuracy capability.
Solution Approach 2:
The patent introduces distributed representations (word embeddings, contextual vectors) as an intermediary layer between the input text and the extraction decision. This intermediary transformation converts diverse named entity representations into a common semantic vector space, where similarity computation becomes straightforward. This mediator approach simplifies the model's task by preprocessing the semantic information before final entity recognition.
3Measurement precision
If distributed representations are used to capture semantic meaning, then the ability to recognize paraphrased entities improves, but the computational requirements and processing time increase
Solution Approach 1:
The patent performs preliminary computation of distributed representations (word embeddings, contextual vectors) before the actual entity extraction process. By pre-computing and storing these semantic representations, the system avoids redundant calculations during extraction, reducing real-time computational requirements. The heavy lifting of semantic encoding is done in advance, allowing faster and more energy-efficient extraction operations.
Solution Approach 2:
The patent computes distributed representations for all words in the text, which is more than strictly necessary for entity extraction alone. However, this excessive computation of full contextual embeddings provides richer semantic information that improves paraphrase recognition accuracy. The system accepts the additional computational cost as a trade-off for significantly enhanced ability to recognize entities with varying representations.
Data Source
AI summary
A learning method to be executed by a computer, the learning method includes when a first input sentence in which a predetermined target is represented by a first named entity is input to a first machine learning model, learning a first parameter of the first machine learning model such that a value output from the first machine learning model approaches correct answer information corresponding to the first input sentence; and when an intermediate representation generated when the first input sentence is input to the first machine learning model and a second input sentence in which the predetermined target is represented by a second named entity are input to a second machine learning model, learning the first parameter and a second parameter of the second machine learning model such that a value output from the second machine learning model approaches correct answer information corresponding to the second input sentence.