Multi-Characteristic NER via Interaction Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Named Entity Recognition (NER) processes require multiple passes through a machine learning model to determine multiple characteristics of text, consuming significant time and computational resources, making them inefficient for large datasets.
Innovation Solution
A method that combines characteristics into a single input for a machine learning model, using techniques like Bidirectional Encoder Representations from Transformers (BERT) to generate interaction embeddings for efficient extraction of multiple characteristics in a single pass, reducing resource consumption and training time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple passes through the machine learning model are used to determine multiple characteristics, then the accuracy and completeness of characteristic extraction is improved, but the time consumption and computational resources increase significantly
Solution Approach 1:
The patent merges multiple characteristic extraction tasks into a single machine learning model inference pass. Instead of running separate passes for each characteristic (e.g., brand, category, color, size), the system combines all characteristics into one unified model that processes the text once and outputs all characteristic labels simultaneously, thereby reducing time consumption while maintaining extraction accuracy
Solution Approach 2:
The machine learning model is designed with multi-functionality to perform multiple NER tasks concurrently. The model is trained to recognize multiple characteristics (brand name, category, color, size) within a single pass, making the system universally applicable to various product attribute extraction tasks without requiring separate specialized models for each characteristic
2Measurement precision
If multiple passes through the machine learning model are used to determine multiple characteristics, then the completeness of characteristic identification is improved, but the computational resources consumed increase significantly
Solution Approach 1:
The system combines multiple characteristic identification tasks into a single computational pass through one machine learning model. This consolidation reduces the total number of model invocations required, thereby decreasing computational resource consumption and energy usage while maintaining complete characteristic identification across all product attributes
3Measurement precision
If multiple passes through the machine learning model are used to determine multiple characteristics, then the thoroughness of text analysis is improved, but the training time and scalability to large datasets deteriorate
Solution Approach 1:
The patent merges multiple training tasks into a single training pass by designing a unified model that learns to identify multiple characteristics simultaneously. This approach reduces the total training iterations and computational overhead required during model development, thereby decreasing training time and improving scalability to large datasets while maintaining thorough text analysis capability
Data Source
AI summary
A machine learning model analyzes text describing an item to determine portions of the text that correspond to multiple characteristics of the item. A first set of embeddings that represent the text describing the item is determined. A second set of embeddings that represent text indicating the characteristics is determined. The second set of embeddings includes a token for each characteristic that is used to indicate tokens that are associated with a particular characteristic. The first set of embeddings and portions of the second set of embeddings for a particular characteristic are used to determine a set of interaction embeddings for each characteristic by determining an element-wise product. These interaction embeddings are analyzed to determine label predictions indicating text that is associated with each characteristic. Text for multiple characteristics may therefore be identified using a single pass rather than multiple passes.