Wookhee Min

Planning-integrated story graph for interactive narratives

The advances in the interactive contents enable users to have a variety of experiences on diverse devices. In particular, two main approaches have been researched to construct digital interactive contents: a) conditional branch techniques and b) planning techniques. Each approach offers its own benefits; the conditional branch techniques allow the user to create tightly-plotted interactive contents; the planning techniques reduce the authors burden to specify every possible connection between contents considering the user input. As an attempt to combine these advantages provided by each technique, this paper discusses an interactive story structure incorporating the planning technique into the conditional branch techniques. Also, we briefly describe PRISM, a framework capable of creating and playing our story structure. We expect that the author can compose well-woven stories which can respond to a wide range of user interaction.

PRISM: A Framework for Authoring Interactive Narratives

The advances in computing technologies enable the computer users to create and share their own stories to the community at large. However, it is still regarded as complicated and laborious to author interactive narratives, where a story adapts as the user interacts with it. In authoring interactive narratives, two main approaches--branching graphs and AI planning--have been significantly used to augment interactivity into conventional linear narratives. Although each approach offers its own possibilities and limitations, few efforts have been made to blend these approaches. This paper describes a framework for authoring interactive narratives that employs an adapted branching narrative structure that also uses planning formalism to enable automated association between nodes. We expect that our work is valuable for non-expert users as well as AI researchers in interactive storytelling who need to create a large quantity of story contents for varied endings for a story.

Personalizing Embedded Assessment Sequences in Narrative-Centered Learning Environments: A Collaborative Filtering Approach

A key challenge posed by narrative-centered learning environments is dynamically tailoring story events to individual students. This paper investigates techniques for sequencing story-centric embedded assessments—a particular type of story event that simultaneously evaluates a student’s knowledge and advances an interactive narrative’s plot—in narrative-centered learning environments. We present an approach for personalizing embedded assessment sequences that is based on collaborative filtering. We examine personalized event sequencing in an edition of the Crystal Island narrative-centered learning environment for literacy education. Using data from a multi-week classroom study with 850 students, we compare two model-based collaborative filtering methods, including probabilistic principal component analysis (PPCA) and non-negative matrix factorization (NMF), to a memory-based baseline model, k-nearest neighbor. Results suggest that PPCA provides the most accurate predictions on average, but NMF provides a better balance between accuracy and run-time efficiency for predicting student performance on story-centric embedded assessment sequences.

DeepStealth: Leveraging Deep Learning Models for Stealth Assessment in Game-Based Learning Environments

A distinctive feature of intelligent game-based learning environments is their capacity for enabling stealth assessment. Stealth assessments gather information about student competencies in a manner that is invisible, and enable drawing valid inferences about student knowledge. We present a framework for stealth assessment that leverages deep learning, a family of machine learning methods that utilize deep artificial neural networks, to infer student competencies in a game-based learning environment for middle grade computational thinking, Engage. Students’ interaction data, collected during a classroom study with Engage, as well as prior knowledge scores, are utilized to train deep networks for predicting students’ post-test performance. Results indicate deep networks that are pre-trained using stacked denoising autoencoders achieve high predictive accuracy, significantly outperforming standard classification techniques such as support vector machines and naive Bayes. The findings suggest that deep learning shows considerable promise for automatically inducing stealth assessment models for intelligent game-based learning environments.

NCSU_SAS_WOOKHEE: A Deep Contextual Long-Short Term Memory Model for Text Normalization

To address the challenges of normalizing online conversational texts prevalent in social media, we propose a contextual long-short term memory (LSTM) recurrent neural network based approach, augmented with a self-generated dictionary normalization technique. Our approach utilizes a sequence of characters as well as the part-of-speech associated with words without harnessing any external lexical resources. This work is evaluated on the English Tweet data set provided by the ACL 2015 W-NUT Normalization of Noisy Text shared task. The results, by achieving second place (F1 score: 81.75%) in the constrained track of the competition, indicate that the proposed LSTM-based approach is a promising tool for normalizing non-standard language.

An Interactive-Content Technique Based Approach to Generating Personalized Advertisement for Privacy Protection

Personalized contents have been getting more attention from industry and academia due to its effective communicative role in product advertisements. However, there exist potential threats to the customers privacy in conventional approaches where a data server containing customer profiles is employed or the customer profiles is required to be sent over the public network. To address this, this paper describes a framework that employs a script-based interactive content technique for privacy protection. We illustrate our approach by a sample scenario.

Interactive Narrative Personalization with Deep Reinforcement Learning.

Data-driven techniques for interactive narrative generation are the subject of growing interest. Reinforcement learning (RL) offers significant potential for devising data-driven interactive narrative generators that tailor players’ story experiences by inducing policies from player interaction logs. A key open question in RL-based interactive narrative generation is how to model complex player interaction patterns to learn effective policies. In this paper we present a deep RL-based interactive narrative generation framework that leverages synthetic data produced by a bipartite simulated player model. Specifically, the framework involves training a set of Q-networks to control adaptable narrative event sequences with long short-term memory network-based simulated players. We investigate the deep RL framework’s performance with an educational interactive narrative, CRYSTAL ISLAND. Results suggest that the deep RL-based narrative generation framework yields effective personalized interactive narratives.

“Thanks Alisha, Keep in Touch”: Gender Effects and Engagement with Virtual Learning Companions

Virtual learning companions have shown significant potential for supporting students. However, there appear to be gender differences in their effectiveness. In order to support all students well, it is important to develop a deeper understanding of the role that student gender plays during interactions with learning companions. This paper reports on a study to explore the impact of student gender and learning companion design. In a three-condition study, we examine middle school students’ interactions in a game-based learning environment that featured one of the following: (1) a learning companion deeply integrated into the narrative of the game; (2) a learning companion whose backstory and personality were not integrated into the narrative but who provided equivalent task support; and (3) no learning companion. The results show that girls were significantly more engaged than boys, particularly with the narrative-integrated agent, while boys reported higher mental demand with that agent. Even when controlling for video game experience and prior knowledge, the gender effects held. These findings contribute to the growing understanding that learning companions must adapt to students’ gender in order to facilitate the most effective learning interactions.

Integrating Real-Time Drawing and Writing Diagnostic Models: An Evidence-Centered Design Framework for Multimodal Science Assessment

Interactively modeling science phenomena enables students to develop rich conceptual understanding of science. While this understanding is often assessed through summative, multiple-choice instruments, science notebooks have been used extensively in elementary and secondary grades as a mechanism to promote and reveal reflection through both drawing and writing. Although each modality has been studied individually, obtaining a comprehensive view of a students conceptual understanding requires analyses of knowledge represented across both modalities. Evidence-centered design ECD provides a framework for diagnostic measurement of data collected from student interactions with complex learning environments. This work utilizes ECD to analyze a corpus of elementary student writings and drawings collected with a digital science notebook. First, a competency model representing the core concepts of each exercise, as well as the curricular unit as a whole, was constructed. Then, evidence models were created to map between student written and drawn artifacts and the shared competency model. Finally, the scores obtained using the evidence models were used to train a deep-learning based model for automated writing assessment, as well as to develop an automated drawing assessment model using topological abstraction. The findings reveal that ECD provides an expressive unified framework for multimodal assessment of science learning with accurate predictions of student learning.

Diagrammatic Student Models: Modeling Student Drawing Performance with Deep Learning

Recent years have seen a growing interest in the role that student drawing can play in learning. Because drawing has been shown to contribute to students’ learning and increase their engagement, developing student models to dynamically support drawing holds significant promise. To this end, we introduce diagrammatic student models, which reason about students’ drawing trajectories to generate a series of predictions about their conceptual knowledge based on their evolving sketches. The diagrammatic student modeling framework utilizes deep learning, a family of machine learning methods based on a deep neural network architecture, to reason about sequences of student drawing actions encoded with temporal and topological features. An evaluation of the deep-learning-based diagrammatic student models suggests that it can predict student performance more accurately and earlier than competitive baseline approaches.