Domain-Trained Embedding for Topic Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing tools struggle to efficiently classify natural language text, particularly in domain-specific contexts, due to lack of training on specific domain language, requiring significant manual labor, and being inefficient in tracking topic growth and identifying emerging issues.
Innovation Solution
A machine learning model is trained using a two-stage approach: supervised learning to classify known topics and unsupervised learning to identify new emerging topics, without requiring full annotation of the training set, and leveraging a base natural language model fine-tuned for domain-specific language.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing natural language processing tools are used without domain-specific training, then general language processing capability is maintained, but proficiency in processing domain-specific feedback deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-training the natural language processing model on domain-specific text data before deployment. The model is trained on gaming-related text to learn domain-specific language patterns, terminology, and expressions, enabling it to accurately process and classify gaming feedback without requiring extensive manual annotation of each feedback item.
2Measurement precision
If manual human annotation is used to achieve desired objectivity and accuracy, then classification quality is improved, but time consumption and labor requirements deteriorate
Solution Approach 1:
The patent applies self-service by implementing a semi-supervised learning approach where the system automatically processes the majority of feedback using the trained model, requiring minimal human intervention. Human annotators only need to review and correct a small sample of classifications for validation, rather than manually annotating every feedback item, thus dramatically reducing annotation time while maintaining accuracy.
3Productivity
If conventional topic classifiers are used, then existing topic classification is possible, but efficiency in tracking topic growth and identifying emerging issues deteriorates
Solution Approach 1:
The patent applies feedback by implementing a continuous learning system where classification results are fed back into the model for ongoing improvement. The system tracks topic trends over time, identifies emerging topics through unsupervised learning on unlabeled data, and uses this information to refine classifications and adapt to new gaming feedback patterns, enabling both efficient tracking and detection of emerging issues.
Data Source
AI summary
A classifier may be trained with less than all datasets manually annotated with labels. A small subset of verbatims may be manually labeled with topic labels as seeds. Data augmentations can be used to acquire seed verbatim sets for known topics and to assign temporary pseudo labels to the rest of the verbatims based on their vector space proximity to the labeled seed verbatims. The training may involve classification epochs during which embeddings are updated with the assumption that the pseudo labels are ground-truth labels. The training may also involve labeling epochs during which the updated embeddings are used to update the vectors corresponding to the verbatims, and pseudo labels are updated based on updated vector coordinates in the vector space. As the training process progresses through the epochs, the embeddings will converge.