Multi-level Neural Tagger for Context-Aware Semantic Tagging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Prior statistical modeling approaches for automated semantic tagging lack context, leading to inaccurate tagging due to independent analysis of words, which results in misinterpretation of text and errors in downstream applications such as search engines and machine translation.
Innovation Solution
A multi-level machine learning model architecture is employed, incorporating deep neural network (DNN)-based feature generation, predictive, and lexicon-based models to generate context-aware features and improve tagging accuracy by avoiding the double looping problem through efficient segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If prior statistical modeling approaches analyze words independently, then computation is simpler and faster, but tagging accuracy deteriorates due to lack of context
Solution Approach 1:
The patent segments the text processing into character-level features and word-level features, allowing independent analysis of characters while maintaining word-level context. This segmentation enables the system to capture sub-word information without requiring full contextual analysis of entire sentences, thus improving accuracy while maintaining computational efficiency.
Solution Approach 2:
The patent introduces a new dimension by analyzing text at the character level in addition to the traditional word level. This multi-level feature extraction approach adds depth to the analysis without requiring proportional increases in computational resources, as character-level features are computed efficiently and integrated with word-level features.
2Measurement precision
If multi-level model architecture incorporates context information, then tagging accuracy is improved, but device complexity increases
Solution Approach 1:
The model architecture is segmented into distinct components: character-level feature extraction, word-level feature extraction, and feature combination layers. This modular segmentation allows each component to be optimized independently and makes the overall complex system more manageable and interpretable.
Solution Approach 2:
The patent employs universal feature extraction mechanisms that operate at multiple levels (character and word) using the same underlying neural network architecture. This multi-functionality allows the system to handle different levels of abstraction with a unified approach, reducing the need for separate specialized components and thereby managing complexity.
3Measurement precision
If deep neural network models are used for feature generation, then context awareness is improved, but computational resources and time increase
Solution Approach 1:
The patent performs preliminary feature extraction at the character level before word-level analysis. By pre-computing character-level features and storing them, the system avoids redundant computations during word-level analysis, thereby reducing overall computational time while maintaining context awareness.
Solution Approach 2:
The system extracts only the most relevant features at each level rather than analyzing all possible features. By selectively focusing on character-level and word-level features that provide the most context information, the system achieves good context awareness with reduced computational overhead compared to exhaustive feature extraction.
Data Source
AI summary
Embodiments of the described technologies are capable of reading a text sequence that include at least one word; extracting model input data from the text sequence, where the model input data includes, for each word of the text sequence, segment data and non-segment data; using a first machine learning model and at least one second machine learning model, generating, for each word of the text sequence, a multi-level feature set; outputting, by a third machine learning model, in response to input to the third machine learning model of the multi-level feature set, a tagged version of the text sequence; executing a search based at least in part on the tagged version of the text sequence.