Search

This report describes a randomized controlled study that compared the personalization of educational content based on neural networks to personalization by human experts. The study was conducted in a graphically rich online learning environment for elementary school mathematics, in which N = 135 fourth- and sixth-grade students learn via mathematical applets. The performance of students who followed the algorithm's recommendations was compared with that of students who followed an a priori sequence constructed by the experts. While the algorithm only considered students' performance on past problems when recommending new problems, the human experts also took into consideration other factors related both to content and to the graphical interface. The findings reveal no significant differences in performance between the two groups, suggesting that the algorithm was as successful in preparing the students as human teachers. Herein, we discuss the different mechanisms used to prepare each of the groups for the learning tasks and highlight the importance of the user interface in that process. Specifically, we find that applets involving supportive interactions, in which students' interactions were intended to help solve the problem but were optional, represented students' preknowledge, while applets entailing required interactions did not. We contribute to the field of personalization in education with new evidence of the advantages of a content sequencing algorithm-based on collaborative filtering ranking and implemented via a neural network--in a graphically rich environment as tested in authentic classrooms.

Promoting creativity is considered an important goal of education, but creativity is notoriously hard to measure. In this article, we make the journey from defining a formal measure of creativity, that is, efficiently computable to applying the measure in a practical domain. The measure is general and relies on core theoretical concepts in creativity theory, namely fluency, flexibility, and originality, integrating with prior cognitive science literature. We adapted the general measure for projects in the popular visual programming language Scratch. We designed a machine learning model for predicting the creativity of Scratch projects, trained and evaluated on human expert creativity assessments in an extensive user study. Our results show that opinions about creativity in Scratch varied widely across experts. The automatic creativity assessment aligned with the assessment of the human experts more than the experts agreed with each other. This is a first step in providing computational models for measuring creativity that can be applied to educational technologies, and to scale up the benefit of creativity education in schools.

This article provides computational and rule-based approaches for detecting confusion that is expressed in students' comments in couse forums. To obtain reliable, ground truth data about which posts exhibit student confusion, we designed a decision tree that facilitates the manual labeling of forum posts by experts. However, manual labeling is costly in time and resources, which limits the amount of data that can be generated using this process. Our strategy for overcoming these limitations was to generate rules for detecting confusion based on student input via hashtags, which reflect the student's affective states. We show that the resulting rules closely align with the ground truth judgement of experts. We next applied these rules to datasets of students' forum posts in a large-scale biology course, thereby automatically generating thousands of labeled instances of "confused posts." Finally, the resulting dataset was used to train a machine learning model for detecting whether students' posts exhibit confusion in the absence of hashtags. In this task, the pretrained language model based on bidirectional encoder representation from transformers (BERT) was able to outperform traditional machine learning models for classifying confusion in posts. This model was also able to generalize and detect student confusion across different offerings of the same course. Ultimately, the use of pretrained language models of this type will provide teachers with better technologies for detecting and alleviating confusion in online discussion forums by leveraging the combined input of teachers and students.

Personalizing the use of educational mathematics applets to fit learners' characteristics poses a great challenge. The present study adopted a unique approach by comparing personalization processes implemented by a machine to those implemented by a human teacher. Given the different affordances--the machine's access to historical log file data, computation and automatization, and the teacher's mathematical knowledge, pedagogical approach and personal acquaintance--the study hypothesized that different considerations would lead to different personalization and learning outcomes. Mathematical applets were assigned to 77 students in the 4th and 5th grades either by an expert teacher or by an algorithm. The assignment took place in a controlled setting in which the teacher was unaware which students were eventually assigned according to her recommendations. The teacher and the machine each recommended an ordered sequence of ten applets per student. The findings suggest that the teacher-assigned group outperformed the machine-assigned group among 5th graders when the applets were sequenced in increasing order of difficulty. In the 4th grade, only the machine recommended a sequence of increasing difficulty and both groups achieved equal performance. The study concludes that in the case of data-driven personalization processes, machines and teachers should learn from each other's affordances and considerations.

Background: This study addresses the suicide risk predicting challenge by exploring the predictive ability of machine learning (ML) models integrated with theory‐driven psychological risk factors in real‐time crisis hotline chats. More importantly, we aimed to understand the specific theory‐driven factors contributing to the ML prediction of suicide risk. Method: The dataset consisted of 17,654 crisis hotline chat sessions classified dichotomously as suicidal or not. We created a suicide risk factors‐based lexicon (SRF), which encompasses language representations of key risk factors derived from the main suicide theories. The ML model (Suicide Risk‐Bert; SR‐BERT) was trained using natural language processing techniques incorporating the SRF lexicon. Results: The results showed that SR‐BERT outperformed the other models. Logistic regression analysis identified several theory‐driven risk factors significantly associated with suicide risk, the prominent ones were hopelessness, history of suicide, self‐harm, and thwarted belongingness. Limitations: The lexicon is limited in its ability to fully encompass all theoretical concepts related to suicide risk, nor to all the language expressions of each concept. The classification of chats was determined by trained but non‐professionals in metal health. Conclusion: This study highlights the potential of how ML models combined with theory‐driven knowledge can improve suicide risk prediction. Our study underscores the importance of hopelessness and thwarted belongingness in suicide risk and thus their role in suicide prevention and intervention. [ABSTRACT FROM AUTHOR]

Objective: Although many suicide risk factors have been identified, there is limited guidance about their relative importance and the combinations of factors that heighten risk among young people experiencing homelessness (YEH). We sought to use decision-tree (DT) analyses to better understand and predict suicidal ideation and suicide attempts among YEH. Method: Using survey and social network methods, we gathered information about 940 YEH and their relationships. We then used a machine learning approach to construct classification and regression tree models to predict suicidal ideation and suicide attempts. Results: Thirteen variables were important correlates in the DT models; this included prominent individual risk factors (e.g., trauma, depression), but more than half were social network factors (e.g., hard drug use). For suicidal ideation, the model had an area under the receiver operating characteristic curve (AUC) value of 0.79, with accuracy of 68%, sensitivity of 48%, and specificity of 73%. For suicide attempt, the model had an AUC value of 0.86, with accuracy of 71%, sensitivity of 68%, and specificity of 72%. Conclusions: Effective suicide prevention programming should target the syndemic that threatens YEH (i.e., co-occurrence of trauma, depression, substance use, and violence), including social norms in their environments. With refinement, our decision trees may be useful for suicide risk screening and guiding targeted intervention. [ABSTRACT FROM AUTHOR]