Neural Network Content Scoring with Spelling Error Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-based systems for automatically scoring constructed responses, such as essays, face challenges in accurately handling spelling errors and require significant time for grading, limiting their efficiency in providing feedback on student understanding and comprehension.
Innovation Solution
The integration of character representations and spell correction within neural network models, specifically using recurrent neural networks (RNNs) and pre-trained transformer-based architectures, to improve the accuracy of content scoring by aggregating numerical vectors through pooling and attention mechanisms, and employing model weight averaging during training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If character-based representations are integrated into neural network models for scoring constructed responses, then the accuracy of content scoring is improved by accounting for spelling errors, but the device complexity increases
Solution Approach 1:
The model segments the input text into character-level units and processes them through separate embedding layers before aggregation. This segmentation allows the model to capture spelling variations at the character level while maintaining word-level semantic understanding, thereby improving scoring accuracy without overwhelming complexity
Solution Approach 2:
The patent combines multiple representation types (character embeddings, word embeddings, and their concatenations) to create a composite input representation. This composite approach integrates morphological information from characters with semantic information from words, achieving higher measurement precision through multi-faceted feature integration
2Productivity
If computer-implemented methods are used to automatically score constructed responses, then the productivity of grading is improved, but the measurement precision deteriorates due to interference from spelling errors
Solution Approach 1:
The patent introduces character-level representations as an intermediary layer between raw text input and word-level semantic processing. This intermediary captures spelling variations and morphological information, acting as a buffer that preserves scoring accuracy while enabling automated processing at scale
Solution Approach 2:
The model dynamically adjusts its processing parameters by selectively attending to character-level features when spelling errors are detected. This parameter adaptation allows the system to maintain high measurement precision across diverse input qualities while preserving automated grading efficiency
3Measurement precision
If model weight averaging is employed during training, then the measurement precision of the scoring model is improved, but the training time increases
Solution Approach 1:
The patent implements weight averaging as a preliminary processing step during the training phase, computing averaged weights across multiple training iterations before final model deployment. This preliminary action embeds robustness against spelling variations into the model structure, achieving improved measurement precision without requiring additional processing time during actual scoring operations
Data Source
AI summary
Systems and methods are disclosed for automatically scoring a constructed response using a neural network. In embodiments, a constructed response received by a processing system may be processed to divide the constructed response into multiple series of word tokens, wherein each word token includes a sequence of characters. The constructed response may be further processed to correct one or more spelling errors. The word tokens may be encoded to generate representation vectors for the constructed response. A set of nonlinear operations may be applied to the plurality of representation vectors in a neural network to generate a single vector output. A set of predetermined network weights may be applied to the vector output of the neural network to generate a scalar output for scoring the constructed response.