bloominzer-0.1

What is the Bloominizer

The bloominer is a fine-tuned version of BERT that classifies questions by the Bloom's Taxonomy level: Knowledge, Comprehension, Application, Analysis, Synthesis, Evaluation.

Tests during training indicate that the Bloominizer is approximately 93% accurate in its classifications, with most misclassifications being for either one level below or above (for instance, it may misclassify a Comprehension question as a Knowledge question, but rately as an Evaluation question).

The Bloominizer has been used for large-scale classification of questions from a corpus. For example, a useful usecase is to run all questions in a long multiple choice exam through the Bloominizer and compute the relative percentages of questions from the six Bloom's levels. This can give you an idea of the approximate cognitive level of the overall exam.

Using in transformers

The Bloominizer is easiest to use through a pipeline. Sample code is below:

import transformers
import torch
from transformers import pipeline
pipe = pipeline("text-classification", model="uw-vta/bloominzer-0.1")
print(pipe("What is a goat?"))

If you run this code, the output should be something like:

[{'label': 'Knowledge', 'score': 0.9993932247161865}]