Phrase Splitting Kernel for Food Text Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current natural language processing technologies fail to accurately split food-related text due to the complexity of compound words and mixed phrases, particularly in the food domain, where existing methods do not effectively utilize multiple information sources to improve splitting accuracy.
Innovation Solution
A computing device with a phrase splitting kernel module that employs multiple phrase splitting models, including a crowd-sourced database, exception database, domain-specific content database, entropy algorithm, and hierarchical rule model, to generate a ranked list of suggested split locations, enabling accurate parsing of digital representations of spoken words into n-grams for food tracking and nutritional information retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional natural language processing methods are used to split compound words, then the processing is simple and fast, but the splitting accuracy is low due to failure to identify proper split locations in food-related compound words and mixed phrases
Solution Approach 1:
The patent segments the compound word splitting task into multiple independent models, each handling specific aspects: n-gram model for common phrases, conditional random field model for contextual dependencies, hidden Markov model for sequential patterns, and maximum entropy model for probabilistic rankings. This segmentation allows each model to specialize in particular patterns while collectively achieving high accuracy.
Solution Approach 2:
The patent merges outputs from multiple independent splitting models through a ranking framework that combines probabilities and scores from each model. The final split locations are determined by aggregating results across all models, leveraging their complementary strengths to achieve superior accuracy compared to any single model alone.
2Measurement precision
If multiple phrase splitting models and information sources are employed to improve splitting accuracy, then the accuracy increases, but the device complexity and computational requirements increase
Solution Approach 1:
The patent implements a universal ranking framework that can accommodate multiple different types of splitting models (n-gram, CRF, HMM, maximum entropy) through a common interface. This framework processes outputs from diverse models uniformly, allowing the system to leverage multiple information sources while maintaining manageable complexity through standardized integration.
Solution Approach 2:
The ranking framework acts as an intermediary layer between the multiple specialized splitting models and the final output. It mediates by standardizing model outputs, combining probabilities, and producing unified split location rankings, thereby managing the complexity of integrating multiple models without requiring complex custom integration logic.
3Measurement precision
If domain-specific databases and multiple algorithms are used to handle food-related text, then the splitting accuracy for food domain improves, but the processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary processing by pre-computing n-gram probabilities and storing domain-specific food phrases in advance. The n-gram model provides pre-calculated probabilities for common food compound words, allowing the system to quickly retrieve and use these pre-computed values during actual splitting operations without re-computing from scratch.
Solution Approach 2:
The system applies partial action by selectively activating and weighting different models based on the specific input characteristics. Not all models are equally weighted or applied to all inputs - the ranking framework can adjust the contribution of each model based on the particular food-related text being processed, optimizing the balance between accuracy and processing efficiency.
Data Source
AI summary
A device capable of splitting user input into phrases is presented. The disclosed device leverages multiple phrase splitting models to generate one or more possible split locations. The possible split locations can be derived based on leveraging multiple phrase splitting models. Each model contributes its suggested split locations to the set of possible split locations according to an implementation of a phrase splitting kernel algorithm that weights each model's suggestions.