response_validation_app

Implements a simple unsupervised method for classifying student short to medium sized responses to questions.

Installation

This was developed in Python 3.6.

It may be installed as a package from the pypi repository, using pip:

pip install response-validator

Development

After cloning the repository, you can install the repo in editable mode, as so:

pip install -e .

Note that this step will download several NLTK corpora, silently, and add them to the deployed tree.

Additional functionality for running algorithm tests, etc. can be enabled by installing additional libraries:

pip install -r requirements.txt

Usage

Development

In order to persist the book vocabulary data between invocations, the Flask server needs the DATA_DIR setting to contain a path pointing to an existing directory. This can be set in several ways.

1. Pass a key-value command line argument:

python -m validator.app DATA_DIR=data

2. set the VALIDATOR_SETTINGS environment variable to the path of a file that contains the DATA_DIR setting:

VALIDATOR_SETTINGS=data/dev.cfg python -m validator.app

Where the contents of dev.cfg is:

DATA_DIR=data

and the directory data exists.

3. use gunicorn, provide arguments to app factory:

gunicorn 'validator.app:create_app(DATA_DIR="data")'

4. Use gunicorn, with an environment variable pointing to a config file:

VALIDATOR_SETTINGS=../data/dev.cfg gunicorn "validator.app:create_app()"

Note that this one can get confusing with relative paths, since flask uses the directory the app is imported from (in this case, validator) as the config path when interpreting environment variables, while paths inside such files will be based on the python current working directory. When in doubt, use full paths:

VALIDATOR_SETTINGS="$PWD/data/dev.cfg" gunicorn "validator.app:create_app()"

Production

The recommended production method for deployment is to use a WSGI compliant server, such as gunicorn:

pip install gunicorn gevent
gunicorn -k gevent -b 5000 "validator.app:create_app(DATA_DIR='/var/lib/validator/data')" 

Ideally, use a socket, and place nginx or other webserver in front of flask, for https termination, if nothing else.

gunicorn -k gevent --bind /run/gunicorn.sock "validator.app:create_app(DATA_DIR='/var/lib/validator/data')"

API

Response Validation

The main route for the app is /validate, which accepts a plaintext response (response) that will be checked. It can also accept a number of optional arguments:

• uid (e.g., '1000@1', default None): This is the uid for the question pertaining to the response. The uid is used to compute domain-specific and module-specific vocabulary to aid in the classification process. Iff the version of the question specified is not available, any version of the same qid (question id without the version, e.g. 1000) will be used.

• remove_stopwords (True or False, default True): Whether or not stopwords (e.g., 'the', 'and', etc) will be removed from the response. This is generally advised since these words carry little predictive value.

• tag_numeric (True, False or auto, default auto): Whether numerical values will be tagged (e.g., 123.7 is tagged with a special 'numeric_type_float' identifier). While there are certainly responses for which this would be helpful, a large amount of student garbage consists of random number pressing which limits the utility of this option. Auto enables a mode that only does numeric tag processing if the question this response pertains to (as fond via the uid above) requires a numeric answer.

• spelling_correction (True, False or auto, default auto): Whether the app will attempt to correct misspellings. This is done by identifying unknown words in the response and seeing if a closely related known word can be substituted. Currently, the app only attempts spelling correction on words of at least 5 characters in length and only considers candidate words that are within an edit distance of 2 from the misspelled word. When running in auto mode, the app will attempt to determine validity without spelling correction. Only if that is not valid, will it attempt to reassess validity with spelling correction.

• spell_correction_max (integer, default 10): Limit spelling corrections applied to this number.

• remove_nonwords (True or False, default True): Words that are not recognized (after possibly attempting spelling correction) are flagged with a special 'nonsense_word' tag. This is done primarily to combat keyboard mashes (e.g., 'asdfljasdfk') that make a large percentage of invalid student responses.

Once the app is running, you can send requests using curl, requests, etc. Here is an example using Python's requests library:

Here an example of how to call things using the Python requests library (assuming that the app is running on the default local development port):

import json
import requests
params = {'response': 'This is my answar to the macromolecules question nitrogenous awerawfsfs'
          'uid': '100@2',
          'remove_stopwords': True,
          'tag_numeric=True': False,
          'spelling_correction': True,
          'remove_nonwords': True}
r = requests.get('http://127.0.0.1:5000/validate', params=params)
print(json.dumps(r.json(), indent=2))
{
  "bad_word_count": 1,
  "common_word_count": 3,
  "computation_time": 0.013212919235229492,
  "domain_word_count": 1,
  "inner_product": 1.5999999999999996,
  "innovation_word_count": 0,
  "intercept": 1,
  "lazy_math_evaluation": true,
  "num_spelling_correction": 2,
  "option_word_count": 0,
  "processed_response": "answer macromolecules question nitrogenous nonsense_word",
  "remove_nonwords": true,
  "remove_stopwords": true,
  "response": "This is my answar to the macromolecules question nitrogenous awerawfsfs",
  "spelling_correction": true,
  "spelling_correction_used": true,
  "stem_word_count": 0,
  "tag_numeric": "auto",
  "tag_numeric_input": "auto",
  "uid_found": true,
  "uid_used": "100@7",
  "valid": true,
  "version": "2.4.0"
}

As you can see from these results, a number of features are taken into account when determining the potential validity of the students response: the words in the response itself, the words from the associated question (stem words) and its answers (option words), the words in the textbook associated with this assignment (domain words), and the words in the textbook whose first appearance is on the page associated with this question (innovation words). Various other features (presence or absence of math, spelling correction, stop word elimination, etc) are also applied. These tests depend on vocabularies being loaded for each exercise.

Service APIs

Here is the /status response for a server started on Oct 15, with a clean install of version 2.4.0, and vocabularies for 5 books loaded:

{
  "datasets": {
    "books": [
      {
        "name": "Biology 2e",
        "vuid": "8d50a0af-948b-4204-a71d-4826cba765b8@15.45"
      },
      {
        "name": "College Physics for AP® Courses",
        "vuid": "8d04a686-d5e8-4798-a27d-c608e4d0e187@26.1"
      },
      {
        "name": "College Physics with Courseware",
        "vuid": "405335a3-7cff-4df2-a9ad-29062a4af261@7.53"
      },
      {
        "name": "Introduction to Sociology 2e",
        "vuid": "02040312-72c8-441e-a685-20e9333f3e1d@10.1"
      },
      {
        "name": "Biology for AP® Courses",
        "vuid": "6c322e32-9fb0-4c4d-a1d7-20c95c5c7af2@18.4"
      }
    ]
  },
  "started": "Tue Oct 15 16:09:23 2019",
  "version": {
    "date": "2019-10-15T14:40:38-0500",
    "dirty": false,
    "error": null,
    "full-revisionid": "463fc5ef4c9d8c37aa600720c8bc814dfa44557c",
    "version": "2.4.0"
  }
}

The datasets at the top list the books that have their vocabularies loaded and available.

Dataset APIs

The following routes all serve JSON formatted representations of the datasets used by the validator to make its validity determinations. Currently, book and exercise vocabularies are available. In the future, this will be expanded to include the weights used to combine the feature values (feature coefficients).

Processing APIs

TODO:

• store feature coefficent sets, return IDs
• additional data APIs for downloading exercise vocabularies and feature cofficient sets
• Currently there is no security for this app (anything can call it). I am not sure how this is usually handled in Tutor but it should not be too difficult to add an api key or similar security measures.
• Depending on UX, we may want to return more granular information about the response rather than a simple valid/non-valid label. We can modify this easily enough as the need arises.