Sparse Factor Analysis for Information Extraction and Fusion
Abstract
Statistical models have been developed to incorporate preferences of each instructor in recommending personalized learning action. Such models can provide feedback to instructors to help them understand their preferences. In addition, novel methods have been proposed to use neural networks to design data-driven question generation models. A new novel criterion is also proposed to evaluate the performance and relevance of such models. Furthermore, we have proposed a method for automatic short answer grading. Finally, stronger generative models have been developed using DRM framework. Over the past year, we have made significant progress on education data processing in five directions: 1) Statistical models for instructor content preference analysis, 2) Data-driven question generation models, 3) Criteria for Neural Question Generation Models, 4) Meta-learning model for automatic short answer grading, 5) Generative models using deep learning.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 21, 2019
- Accession Number
- AD1086043
Entities
People
- Richard G. Baraniuk
Organizations
- Rice University