Differentiated instruction plays a critical role in today's inclusive classrooms to meet the diverse needs of individual students for allowing all students to access the same classroom curriculum by providing different entry points and learning tasks that are tailored to students' needs. The proposed research is to construct a differentiated instructional system of mathematical word problem solving with the following functionalities. First, the system maintains a pool of instructional materials generated by pre-defined templates or shared from students/teachers; features such as readability and the noise level of irrelevant information will be automatically extracted from instructional materials by proposed statistical natural language processing techniques. Second, the system provides computer-assisted instruction to train students' abilities for analyzing and solving mathematical word problems. Third, it enables formative evaluation to monitor students' progress. Fourth, the system provides the recommendation of differentiated instructional materials for a specific student by utilizing a student performance-driven recommendation algorithm. The proposed work includes cutting-edge research for computer science techniques. A joint statistical learning algorithm will be designed for identifying levels of readability, word difficulty, and syntactic complexity of available instructional materials in a unified framework. A relational learning algorithm will be designed for identifying sentences with relevant information from sentences with irrelevant information in a mathematical word problem.
