Flask webapp with NL model integration

Requirements

• flask
• scikit-learn
• nltk

---

Usage

To run the app use

python app.py

Optionally you may add arguments to the command

`--bind`:    Binds the server to the default port (5000) on the loopback IP
             so that it is visible over a local network
`--retrain`: Sets the server to retrain its model every 10 job entries.
             * This is disabled by default as it takes a little bit of
               extra time to retrain a model and Flask has no good way
               to reactively disable buttons.
             * This will also affect the accuracy of the classifier as new
               categories will have much less data and would not be likely
               to be predicted until more data a retrieved.

Structure

Directory Structure

This web app runs under the assumption of job information being in the directory format of /<data | user>/<category>/<job info> and that all web indexes are unique. The data directory contains the same data from milestone 1 to populate the job board. The user directory is created as needed when a job post is made. A job listing must be in the specific format (newline separated, Title: and Description: keys).

When posting a job, users may click on the classify button to classify their job entry. This requires a title and description. Attempting to submit without a category will lead to an automatic classification as the category is pivotal to the data structure.

Users may ignore the classifier and enter custom categories. This will be reflected in the categories navbar and the endpoint will automatically update.

If the model is set to retrain, the vectorizer and model will be replaced when the update occurs.

If you want the original model, this can be found in the defaults directory. It is also safe to delete the user directory to reset the job ads and categories.

Code Structure

The app uses a controller to handle the job data. This is so the main app file is less messy due to the high amount of mapping parameters and the logic of the app is contained together. The classification is handled by the classifier which handles the model, which the controller will make calls to. The classifier handles the data processing so that it can be used by the model.