Candidate Character Ranking Using Transition Weights on Mobile Input
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing input method technologies face challenges in accurately predicting user input due to incomplete coverage and high training and maintenance costs, especially when deploying on mobile devices, and lack of pertinence in adjusting model parameters.
Innovation Solution
A method and apparatus that determine character transition weights and generate sort weights by combining user input probabilities with pre-acquired basic weights, using a discriminative model trained with sample weights to adjust parameters, reducing the need for massive parameters and improving accuracy on mobile devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a discriminative model with massive parameters is used to improve prediction accuracy, then the accuracy of candidate character string presentation is improved, but the device complexity and deployment cost increase significantly
Solution Approach 1:
The patent segments the weight determination process into two independent components: character transition weights (capturing sequential dependencies) and basic weights (capturing static character relationships). This segmentation allows each component to be computed independently with simpler algorithms, avoiding the need for massive parameter models while maintaining comprehensive prediction accuracy.
Solution Approach 2:
The patent extracts and separates the sequential dependency information into character transition weights, which are computed independently from static basic weights. This extraction allows the system to capture temporal patterns without requiring complex recurrent neural networks or transformers, reducing model complexity while preserving predictive power.
2Measurement precision
If comprehensive error correction strategies are implemented to improve prediction accuracy, then the coverage of input scenarios is improved, but the training and maintenance costs increase
Solution Approach 1:
The character transition weights are computed using self-service mechanisms based on observed character sequences in the training data. The system automatically learns transition probabilities from raw input data without requiring complex training procedures, expert intervention, or expensive hyperparameter tuning, thereby reducing training and maintenance costs while improving coverage.
3Adaptability or versatility
If general-purpose input methods are used to improve adaptability, then the versatility of input scenarios is improved, but the pertinence of error correction decreases
Solution Approach 1:
The patent applies local quality by computing character transition weights specific to each character position and context in the input sequence. Instead of using a single general-purpose model, the system adapts its prediction to local characteristics of the input string, improving error correction pertinence while maintaining versatility across different input scenarios.
Data Source
AI summary
A method and an apparatus for presenting a candidate character string and a method and an apparatus for training a discriminative model are provided. The method for presenting a candidate character string may include: acquiring a character string of alphabetic words inputted by a user; determining a character transition weight of the character string, the character transition weight being used to characterize a transition probability of a character; generating a sort weight corresponding to the character string based on a pre-acquired basic weight matching the character string and the character transition weight; and selecting, according to an order indicated by the sort weight, at least two candidate character strings matching the character string from a pre-acquired candidate character string set for presentation.