Educational content labeled with proper knowledge components (KCs) are particularly useful to teachers or content organizers. However, manually labeling educational content is labor intensive and error-prone. To address this challenge, prior research proposed machine learning based solutions to auto-label educational content with limited success. In this work, we significantly improve prior research by (1) expanding the input types to include KC descriptions, instructional video titles, and problem descriptions (i.e., three types of prediction task), (2) doubling the granularity of the prediction from 198 to 385 KC labels (i.e., more practical setting but much harder multinomial classification problem), (3) improving the prediction accuracies by 0.5–2.3% using Task-adaptive Pre-trained BERT, outperforming six baselines, and (4) proposing a simple evaluation measure by which we can recover 56–73% of mispredicted KC labels. All codes and data sets in the experiments are available at: https://github.com/tbs17/TAPT-BERT
Keywords
more »
« less
Classifying Math Knowledge Components via Task-Adaptive Pre-Trained BERT.
Educational content labeled with proper knowledge components (KCs) are particularly useful to teachers or content organizers. However, manually labeling educational content is labor intensive and error-prone. To address this challenge, prior research proposed machine learning based solutions to auto-label educational content with limited success. In this work, we significantly improve prior research by (1) expanding the input types to include KC descriptions, instructional video titles, and problem descriptions (i.e., three types of prediction task), (2) doubling the granularity of the prediction from 198 to 385 KC labels (i.e., more practical setting but much harder multinomial classification problem), (3) improving the prediction accuracies by 0.5–2.3% using Task-adaptive Pre-trained BERT, outperforming six baselines, and (4) proposing a simple evaluation measure by which we can recover 56–73% of mispredicted KC labels. All codes and data sets in the experiments are available at: https://github.com/tbs17/TAPT-BERT
more »
« less
- PAR ID:
- 10285770
- Date Published:
- Journal Name:
- Artificial Intelligence in Education
- Page Range / eLocation ID:
- 408- 419
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Since the introduction of the original BERT (i.e., BASE BERT), researchers have developed various customized BERT models with improved performance for specific domains and tasks by exploiting the benefits of transfer learning. Due to the nature of mathematical texts, which often use domain specific vocabulary along with equations and math symbols, we posit that the development of a new BERT model for mathematics would be useful for many mathematical downstream tasks. In this resource paper, we introduce our multi-institutional effort (i.e., two learning platforms and three academic institutions in the US) toward this need: MathBERT, a model created by pre-training the BASE BERT model on a large mathematical corpus ranging from pre-kindergarten (pre-k), to high-school, to college graduate level mathematical content. In addition, we select three general NLP tasks that are often used in mathematics education: prediction of knowledge component, auto-grading open-ended Q&A, and knowledge tracing, to demonstrate the superiority of MathBERT over BASE BERT. Our experiments show that MathBERT outperforms prior best methods by 1.2-22% and BASE BERT by 2-8% on these tasks. In addition, we build a mathematics specific vocabulary ‘mathVocab’ to train with MathBERT. We discover that MathBERT pre-trained with ‘mathVocab’ outperforms MathBERT trained with the BASE BERT vocabulary (i.e., ‘origVocab’). MathBERT is currently being adopted at the participated leaning platforms: Stride, Inc, a commercial educational resource provider, and ASSISTments.org, a free online educational platform. We release MathBERT for public usage at: https://github.com/tbs17/MathBERT.more » « less
-
Malicious cyber activities impose substantial costs on the U.S. economy and global markets. Cyber-criminals often use information-sharing social media platforms such as paste sites (e.g., Pastebin) to share vast amounts of plain text content related to Personally Identifiable Information (PII), credit card numbers, exploit code, malware, and other sensitive content. Paste sites can provide targeted Cyber Threat Intelligence (CTI) about potential threats and prior breaches. In this research, we propose a novel Bidirectional Encoder Representation from Transformers (BERT) with Latent Dirichlet Allocation (LDA) model to categorize pastes automatically. Our proposed BERTLDA model leverages a neural network transformer architecture to capture sequential dependencies when representing each sentence in a paste. BERT-LDA replaces the Bag-of-Words (BoW) approach in the conventional LDA with a Bag-of-Labels (BoL) that encompasses class labels at the sequence level. We compared the performance of the proposed BERT-LDA against the conventional LDA and BERT-LDA variants (e.g., GPT2-LDA) on 4,254,453 pastes from three paste sites. Experiment results indicate that the proposed BERT-LDA outperformed the standard LDA and each BERT-LDA variant in terms of perplexity on each paste site. Results of our BERTLDA case study suggest that significant content relating to hacker community activities, malicious code, network and website vulnerabilities, and PII are shared on paste sites. The insights provided by this study could be used by organizations to proactively mitigate potential damage on their infrastructure.more » « less
-
Research spanning nearly a century has found that math plays an important role in the learning of chemistry. Here, we use a large dataset of student interactions with online courseware to investigate the details of this link between math and chemistry. The activities in the courseware are labeled against a list of knowledge components (KCs) covered by the content, and student interactions are tracked over a full semester of general chemistry at a range of institutions. Logistic regression is used to model student performance as a function of the number of opportunities a student has taken to engage with a particular KC. This regression analysis generates estimates of both the initial knowledge and the learning rate for each student and each KC. Consistent with results from other domains, the initial knowledge varies substantially across students, but the learning rate is nearly the same for all students. The role of math is investigated by labeling each KC with the level of math involved. The overwhelming result from regressions based on these labels is that only the initial knowledge varies strongly across students and across the level of math involved in a particular topic. The student learning rate is nearly independent of both the level of math involved in a KC and the prior mathematical preparation of an individual student. The observation that the primary challenge for students lies in initial knowledge, rather than learning rate, may have implications for course and curriculum design.more » « less
-
Machine learning with missing data has been approached in two different ways, including feature imputation where missing feature values are estimated based on observed values and label prediction where downstream labels are learned directly from incomplete data. However, existing imputation models tend to have strong prior assumptions and cannot learn from downstream tasks, while models targeting label prediction often involve heuristics and can encounter scalability issues. Here we propose GRAPE, a graph-based framework for feature imputation as well as label prediction. GRAPE tackles the missing data problem using a graph representation, where the observations and features are viewed as two types of nodes in a bipartite graph, and the observed feature values as edges. Under the GRAPE framework, the feature imputation is formulated as an edge-level prediction task and the label prediction as a node-level prediction task. These tasks are then solved with Graph Neural Networks. Experimental results on nine benchmark datasets show that GRAPE yields 20% lower mean absolute error for imputation tasks and 10% lower for label prediction tasks, compared with existing state-of-the-art methods.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1007/978-3-030-78292-4_33
