Interactive Predictor Learning with Interest-Score Label Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional predictor learning requires large amounts of teacher data and labels, but the effectiveness of these data and labels is unclear, leading to inefficient and time-consuming learning processes.
Innovation Solution
A predictor interactive learning system that uses an interest score calculation unit to select teacher data and labels based on statistical analysis, an interactive learning frame to extract relevant data, and a question-response unit to confirm label accuracy, reducing the need for extensive data and labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of teacher data and teacher labels are used for learning, then the prediction accuracy of the predictor is improved, but the learning time becomes excessively long and the process becomes inefficient
Solution Approach 1:
The patent extracts only the most effective teacher data and teacher labels from the corpus based on interest scores, rather than using all available data. The interest score calculation unit identifies words that will provide the most learning benefit, and the interactive learning frame unit extracts only those specific words as teacher data, eliminating unnecessary data from the learning process
Solution Approach 2:
The patent changes the parameter of data selection from using all corpus data to using selectively chosen data based on interest scores. The system dynamically adjusts which data points are used for learning by calculating interest scores that reflect the potential learning effectiveness of each word, thereby optimizing the data subset used for training
2Measurement precision
If a large amount of teacher data and teacher labels are used for learning, then the prediction accuracy of the predictor is improved, but the learning efficiency deteriorates
Solution Approach 1:
The patent implements feedback through the interest score calculation mechanism. The system calculates interest scores based on the current state of the predictor and the corpus, uses this feedback to select the next batch of teacher data and labels, and continuously refines the selection based on learning progress. This feedback loop ensures that only data contributing to improved prediction accuracy is used
Solution Approach 2:
The system performs self-service by automatically identifying and selecting its own most effective training data through the interest score calculation and interactive learning frame units, rather than relying on manual data selection or using all available data indiscriminately
3Measurement precision
If much teacher data and many teacher labels are used, then the prediction accuracy is improved, but it is unclear whether each data point is effective for learning
Solution Approach 1:
The patent introduces an intermediary mechanism in the form of the interest score calculation unit, which acts as a mediator between the corpus and the learning process. This intermediary calculates and evaluates the potential effectiveness of each word as teacher data before it is used for learning, providing information about data effectiveness that would otherwise be lost
Data Source
AI summary
A predictor interactive learning system of the present invention includes a machine learning unit configured to perform machine learning of a predictor that outputs a predicted value indicating a likelihood of being a predetermined intrinsic expression, by using teacher data and teacher labels, an interest score calculation unit configured to obtain an interest score according to statistical data of a corresponding word in a corpus including the predicted value of the predictor for each of words of the corpus, an interactive learning frame unit configured to extract the word serving as the teacher data used in next learning of the predictor according to the interest score, and a question-response unit configured to output a question of whether the extracted teacher data is an intrinsic expression of which the likelihood is predicted by the predictor, and to acquire a teacher label corresponding to the teacher data, as a response to the question, in which the machine learning unit performs machine learning of the predictor using teacher data extracted by a teacher word extraction unit and a teacher label acquired by an interaction unit.