Advancement Auto-Assessment of Students Knowledge States from Natural Language Input

Knowledge Assessment is a key element in adaptive instructional systems and in particular in Intelligent Tutoring Systems because fully adaptive tutoring presupposes accurate assessment. However, this is a challenging research problem as numerous factors affect students’ knowledge state estimation such as the difficulty level of the problem, time spent in solving the problem, etc. In this research work, we tackle this research problem from three perspectives: assessing the prior knowledge of students, assessing the natural language short and long students’ responses, and knowledge tracing.Prior knowledge assessment is an important component of knowledge assessment as it facilitates the adaptation of the instruction from the very beginning, i.e., when the student starts interacting with the (computer) tutor. Grouping students into groups with similar mental models and patterns of prior level of knowledge allows the system to select the right level of scaffolding for each group of students. While not adapting instruction to each individual learner, the advantage of adapting to groups of students based on a limited number of prior knowledge levels has the advantage of decreasing the authoring costs of the tutoring system. To achieve this goal of identifying or clustering students based on their prior knowledge, we have employed effective clustering algorithms. Automatically assessing open-ended student responses is another challenging aspect of knowledge assessment in ITSs. In dialogue-based ITSs, the main interaction between the learner and the system is natural language dialogue in which students freely respond to various system prompts or initiate dialogue moves in mixed-initiative dialogue systems. Assessing freely generated student responses in such contexts is challenging as students can express the same idea in different ways owing to different individual style preferences and varied individual cognitive abilities. To address this challenging task, we have proposed several novel deep learning models as they are capable to capture rich high-level semantic features of text. Knowledge tracing (KT) is an important type of knowledge assessment which consists of tracking students’ mastery of knowledge over time and predicting their future performances. Despite the state-of-the-art results of deep learning in this task, it has many limitations. For instance, most of the proposed methods ignore pertinent information (e.g., Prior knowledge) that can enhance the knowledge tracing capability and performance. Working toward this objective, we have proposed a generic deep learning framework that accounts for the engagement level of students, the difficulty of questions and the semantics of the questions and uses a novel times series model called Temporal Convolutional Network for future performance prediction. The advanced auto-assessment methods presented in this dissertation should enable better ways to estimate learner’s knowledge states and in turn the adaptive scaffolding those systems can provide which in turn should lead to more effective tutoring and better learning gains for students. Furthermore, the proposed method should enable more scalable development and deployment of ITSs across topics and domains for the benefit of all learners of all ages and backgrounds