animate-object/comprende

PoC generating recall/comprehensions tests for news articles

¿Comprende?

What if you had to read an article before you commented on it? Generating efficient reading recall tests and basic understanding (topics, authorship, etc).

Installation

Pre-reqs

• > python 3.8 (might work with lower pythons, ymmv)
• pipenv

1. clone the repo

2. install pythond deps

pipenv install

3. download nltk data, edit NLTK_DATA var in .env to configure path

pipenv run init_corpora

Run tests

pipenv run test

CLI

This ships with a cli module, easily accessible via pipenv:

➜ pipenv run cli -h
Loading .env environment variables…
usage: cli.py [-h] [-l] [-m {important_words} [{important_words} ...]] [-c COUNT] [-d] file

positional arguments:
  file                  the file to analyze

optional arguments:
  -h, --help            show this help message and exit
  -l, --local           look for the file in comprende/tests/data
  -m {important_words} [{important_words} ...], --modules {important_words} [{important_words} ...]
                        specify which question modules to run for this file
  -c COUNT, --count COUNT
                        specify how many questions to generate
  -d, --debug           enable additional debug output

example usage:

➜ pipenv run cli proprietary/nyt/one_district -dl -c2 | jq
Loading .env environment variables…
[
  {
    "prompt": "Which of these key phrases was mentioned in this document?",
    "correct_options": [
      "high-needs students"
    ],
    "module_name": "important_words",
    "additional_option": [],
    "subtype": "least",
    "debug": {
      "freq": 1
    }
  },
  {
    "prompt": "Which of these key phrases was mentioned often in this document?",
    "correct_options": [
      "mr. miyashiro"
    ],
    "module_name": "important_words",
    "additional_option": [],
    "subtype": "most",
    "debug": {
      "freq": 10
    }
  }
]