Showing 1–2 of 2 results for author: Klebanov, B B
-
Anna Karenina Strikes Again: Pre-Trained LLM Embeddings May Favor High-Performing Learners
Authors:
Abigail Gurin Schleifer,
Beata Beigman Klebanov,
Moriah Ariely,
Giora Alexandron
Abstract:
Unsupervised clustering of student responses to open-ended questions into behavioral and cognitive profiles using pre-trained LLM embeddings is an emerging technique, but little is known about how well this captures pedagogically meaningful information. We investigate this in the context of student responses to open-ended questions in biology, which were previously analyzed and clustered by expert…
▽ More
Unsupervised clustering of student responses to open-ended questions into behavioral and cognitive profiles using pre-trained LLM embeddings is an emerging technique, but little is known about how well this captures pedagogically meaningful information. We investigate this in the context of student responses to open-ended questions in biology, which were previously analyzed and clustered by experts into theory-driven Knowledge Profiles (KPs). Comparing these KPs to ones discovered by purely data-driven clustering techniques, we report poor discoverability of most KPs, except for the ones including the correct answers. We trace this "discoverability bias" to the representations of KPs in the pre-trained LLM embeddings space.
△ Less
Submitted 6 June, 2024;
originally announced June 2024.
-
An Exploratory Study of Argumentative Writing by Young Students: A Transformer-based Approach
Authors:
Debanjan Ghosh,
Beata Beigman Klebanov,
Yi Song
Abstract:
We present a computational exploration of argument critique writing by young students. Middle school students were asked to criticize an argument presented in the prompt, focusing on identifying and explaining the reasoning flaws. This task resembles an established college-level argument critique task. Lexical and discourse features that utilize detailed domain knowledge to identify critiques exis…
▽ More
We present a computational exploration of argument critique writing by young students. Middle school students were asked to criticize an argument presented in the prompt, focusing on identifying and explaining the reasoning flaws. This task resembles an established college-level argument critique task. Lexical and discourse features that utilize detailed domain knowledge to identify critiques exist for the college task but do not perform well on the young students data. Instead, transformer-based architecture (e.g., BERT) fine-tuned on a large corpus of critique essays from the college task performs much better (over 20% improvement in F1 score). Analysis of the performance of various configurations of the system suggests that while children's writing does not exhibit the standard discourse structure of an argumentative essay, it does share basic local sequential structures with the more mature writers.
△ Less
Submitted 17 June, 2020;
originally announced June 2020.
