Novelty modeling in unstructured text data has been one of the research concentrations within the Natural Language Processing (NLP) community over the past few years. Effective novelty models can play a key role in providing relevant and interesting content to the users which is the central goal in many applications including educational recommender systems. Computational models of novelty provide formal representations for evaluating and generating creative artifacts in creativity and cognition research. Advances in Natural Language Processing provide new approaches to evaluating computational novelty in unstructured text to be applied in multiple cross-disciplinary research areas including Artificial Intelligence, Education, and Human-Computer Interaction. The problem of novelty measurement in the domain of text has been investigated from different perspectives for different types of textual data. A less examined approach for modeling novelty in unstructured text documents is using Topic Models as the data representation method for gauging computational novelty in research publications. Topic Modeling is a machine learning approach that derives the main themes of a corpus of text documents and represents how they relate. Representing documents with Topic Models has properties that facilitate using various methods for modeling novelty in research publications and also learning materials to be recommended in educational recommender systems. In this dissertation, we first define a framework for characterizing computational models of novelty that is independent of the type of data in the items. This framework enables an exploration and comparison of existing approaches to computational novelty. We then describe and explore an educational recommender system called Pique that applies computational models of novelty to encourage curiosity and self-directed learning by presenting a sequence of learning materials that are both novel and personalized to learners’ interests. We demonstrate how our computational novelty framework can be applied as the AI component of (educational) recommender systems like Pique, and the usefulness of applying computational models of novelty in educational recommender systems to encourage students’ curiosity for expanding their knowledge. We report the student experiences with Pique in four university courses that applied Pique. Based on a qualitative analysis, the students’ experience with Pique encouraged their curiosity and led them to unexpected topics in their projects. We then develop two computational approaches to measuring novelty in research publications using Topic Modeling results and demonstrate these models on a database of research publication abstracts from the ACM CHI Symposia. We analyze and describe how the two novelty models differ in the results and interpretation of novelty. Finally, we compare the computational models of novelty based on Topic Models with human perception of novelty by running a study and recruiting experts in the domain of our dataset (HCI) and report on the results. The qualitative analysis of the results suggested that the novelty model based on topic co-occurrence is slightly closer to human perception of novelty compared to the novelty model based on topics similarity. We also found that the criteria for evaluating novelty of a research publication in humans may not be a complete match with the computational models suggesting these two could complement each other.