Full scrape
The repository contains the full professor scrape with a 20-review cap per professor. That layer is used for coverage and concentration diagnostics.
Interactive report
Ivy League sentiment and reporting bias are different problems
This site combines a full Ivy League professor scrape with a professor-balanced transformer sentiment sample. The main conclusion is not just which school scores highest. It is that review concentration materially changes how reliable those rankings are.
School view
Balanced sentiment across schools
This ranking uses one recent review per reviewed professor. It is the cleaner comparison when the goal is to reduce dominance from a small number of highly reviewed professors.
Balanced sentiment by school
Hover for exact values and coverage details.
Sentiment versus professor rating
Marker size scales with balanced review count.
School summary table
Sorted by balanced sentiment.
| School | Sentiment | Positive | Balanced Reviews | Professors | Detail |
|---|
Bias diagnostics
The strongest signal is review concentration
The full scrape reveals how much of each school’s public profile comes from a narrow set of professors. That is the reporting-bias layer the balanced NLP sample is trying to correct for.
Top 10% review share
Higher means a small professor subset dominates the review pool.
Coverage versus concentration
Lower zero-review share and lower concentration is the healthier quadrant.
Full-scrape coverage table
These figures come from the full professor scrape, not the balanced NLP subset.
| School | Full Reviews | Median Reviews / Professor | Zero-Review Professors | Top 10% Share |
|---|
Department view
Interactive department comparisons
Department results are sensitive to coverage. Use the school filter and minimum review threshold to focus on segments with enough data to be interpretable.
Highest-sentiment departments
Filtered interactively by school and minimum review count.
Lowest-sentiment departments
Negative side of the same filtered population.
Top departments table
| School | Department | Sentiment | Reviews |
|---|
Bottom departments table
| School | Department | Sentiment | Reviews |
|---|
Downloads
Direct data exports
These links point to the generated CSV outputs copied into the deployable site bundle.
Raw site data JSON
Combined school, department, and bias data used by the interactive site.
School summary CSV
Balanced model-level school comparison.
Department summary CSV
Department-level sentiment and coverage.
Model-scored reviews CSV
Balanced professor-level review sample with sentiment labels.
Full professors CSV
All scraped professors across the Ivy League set.
Full reviews CSV
All scraped reviews from the capped full scrape.
Method
Interpretation constraints
Balanced NLP sample
The sentiment ranking uses one review per reviewed professor. That is a methodological correction, not a convenience sample.
Model
Sentiment scores were generated with distilbert-base-uncased-finetuned-sst-2-english and mapped onto a signed scale from -1 to +1.
What not to infer
This does not measure educational quality directly. It measures review text sentiment and the shape of who gets reviewed.
Static figures
Supplemental exports
