We develop models to classify desirable reasoning revisions in argumentative writing. We explore two approaches – multi-task learning and transfer learning – to take advantage of auxiliary sources of revision data for similar tasks. Results of intrinsic and extrinsic evaluations show that both approaches can indeed improve classifier performance over baselines. While multi-task learning shows that training on different sources of data at the same time may improve performance, transfer-learning better represents the relationship between the data. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Providing timely assistance to students in intelligent tutoring systems is a challenging research problem. In this study, we aim to address this problem by determining when to provide proactive help with autoencoder based feature learning and a deep reinforcement learning (DRL) model. To increase generalizability, we only use domain-independent features for the policy. The proposed pedagogical policy provides next-step proactive hints based on the prediction of the DRL model. We conduct a study to examine the effectiveness of the new policy in an intelligent logic tutor. Our findings provide insight into the use of DRL policies utilizing autoencoder based feature learning to determine when to provide proactive help to students. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Real-time feedback is very important, yet challenging to provide for free-text learner contributions in Technology-Enhanced Learning. We study whether a generic NLP pipeline can identify completeness features of learner ideas during security training. We apply PoS Tagging and Dependency Parsing on contextualised short texts, collected within a dedicated learning environment and we compare the results to an expert-annotated ground truth. We scan these contributions for the absence of responsible stakeholder (who) or featured action (how). A total of 1174 contributions in two security domains were analysed. We report precision on who (PPV= 0.929 ) and on how (PPV= 0.691 ). We consider the first result to be sufficient to provide real-time formative feedback for the case of absent who. Our results suggest that for the purposes of providing feedback in free input problem-solving exercises, generic transformer pipelines without fine-tuning can achieve good performance on stakeholder identification. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Large language models have recently been able to perform well in a wide variety of circumstances. In this work, we explore the possibility of large language models, specifically GPT-3, to write explanations for middle-school mathematics problems, with the goal of eventually using this process to rapidly generate explanations for the mathematics problems of new curricula as they emerge, shortening the time to integrate new curricula into online learning platforms. To generate explanations, two approaches were taken. The first approach attempted to summarize the salient advice in tutoring chat logs between students and live tutors. The second approach attempted to generate explanations using few-shot learning from explanations written by teachers for similar mathematics problems. After explanations were generated, a survey was used to compare their quality to that of explanations written by teachers. We test our methodology using the GPT-3 language model. Ultimately, the synthetic explanations were unable to outperform teacher written explanations. In the future more powerful large language models may be employed, and GPT-3 may still be effective as a tool to augment teachers’ process for writing explanations, rather than as a tool to replace them. The explanations, survey results, analysis code, and a dataset of tutoring chat logs are all available at https://osf.io/wh5n9/. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

While many learning and assessment models focus on the binary correctness of student responses, previous studies have shown that having access to extra information—such as the time it takes students to respond to a question—can improve the performance of these models. As much of the previous work in this area has focused on knowledge tracing and next answer correctness, in this study we take a different approach and analyze the relationship between these extra types of information and the overall knowledge of the student, as measured by the end result of an adaptive assessment. In addition to looking at student response times, we investigate the benefit of having detailed information on the responses in the form of answer feedback tags from the adaptive assessment system. After using feature embeddings to encode the information from these feedback tags, we build several models and perform a feature importance analysis to compare the relative significance of these different variables. Although it appears that the response time variable does contain useful information, the answer feedback tags are ultimately much more important to the models. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.