Column Heading Detection Using Word and Character Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated systems for assigning column headings in data lack accuracy, particularly for unknown column-heading types, hindering the efficient processing of data without context.
Innovation Solution
Utilizing masked language models trained on diverse data sets to generate contextual entity and character embeddings, which are then merged and classified using a neural network to predict missing column headings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated systems use traditional machine learning approaches (Sherlock, Sato) for column heading prediction, then processing speed increases, but accuracy deteriorates especially for unknown column-heading types
Solution Approach 1:
The patent segments the column heading prediction task into multiple specialized models: a known-type predictor for standard column headings and an unknown-type predictor for novel column headings. This segmentation allows each model to specialize in its domain, improving overall accuracy while maintaining automated processing speed.
Solution Approach 2:
The patent changes the operational parameters of the prediction system by switching between different prediction models based on the detected type of column heading. When a column heading is identified as known-type, the traditional ML model is used; when identified as unknown-type, the specialized unknown-type predictor is activated, optimizing accuracy for each category.
2Measurement precision
If human reviewers manually assign column headings to data, then prediction accuracy improves, but processing time increases and automation level decreases
Solution Approach 1:
The patent implements a self-service automated system that detects and predicts column headings without human intervention. The dual-model approach enables the system to handle both known and unknown column heading types autonomously, eliminating the need for manual review while maintaining high accuracy.
Solution Approach 2:
The patent introduces an intermediary detection mechanism that automatically identifies whether a column heading belongs to the known-type or unknown-type category. This intermediary step routes predictions to the appropriate model, enabling fully automated processing with accuracy comparable to or exceeding manual review.
3Device complexity
If automated systems are designed to handle only known column-heading types, then system complexity is reduced, but adaptability deteriorates when encountering unknown types
Solution Approach 1:
The patent creates a universal prediction system that handles both known and unknown column heading types through a multi-functional architecture. The system includes a type detection module that identifies whether column headings are known or unknown, and routes to appropriate prediction models, enabling the system to adapt to various input types without requiring complete redesign.
Solution Approach 2:
The patent performs preliminary classification of column headings into known-type and unknown-type categories before prediction. This preliminary action enables the system to prepare the appropriate prediction model in advance, improving adaptability to unknown types while maintaining manageable system complexity through structured workflow.
Data Source
AI summary
A system and method utilizes masked language models in order to provide data-type detection, such as (but not limited to) prediction of columnar headings. Two masked language models are pre-trained on example columnar text. One model predicts missing data at the entity level (e.g., masked entity names that may be made up of whole words), while the other predicts missing data at the character level (e.g., masked individual characters). The table with missing column headings is fed into both models, and the output is contextual word embeddings and contextual character embeddings. These results are merged, and then fed into a neural network classifier to then predict the column names.