Statistics Student Reports with Rhetorical Labels and sBERT Embeddings

1.0 GB of labeled sentences from student-written statistics reports from Carnegie Mellon University. The dataset includes sequences of sentences labeled for rhetorical purpose using Llama and 384-dimensional sBERT embeddings for the original text. It was authored by Margaret Ellingwood and last updated in May 2026.

Use Cases

• Modeling rhetorical structure in academic writing based on labeled sentence sequences.
• Training or evaluating sequence models like Markov mixture models on educational text.
• Using sBERT embeddings for downstream NLP tasks on student writing.
• Analyzing patterns in the rhetorical moves of statistics reports.

Strengths

• Includes 384-dimensional sBERT embeddings for each text segment.
• Sentence-level rhetorical labels were generated using the Llama model.
• Dataset size is 1.0 GB, indicating a medium-scale collection.

Limitations

• The full text of the reports is not shared for privacy reasons, limiting direct content analysis.
• Row count and column-level documentation are unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific student population and courses at Carnegie Mellon University.

Provenance

Source

Student-written reports from statistics courses at Carnegie Mellon University.

Collection Method

Reports were split into sentences and labeled for rhetorical purpose using Llama; sBERT embeddings were generated.

Freshness

Last updated 2026-05-19 16:39:02.