Classification of job positions by area

This is a project to classify job positions using machine learning, more specifically, supervised learning. The main goal is to get a classifier that receives a job position in the form of a sentence, written in natural language, for example CEO and Founder and returns the job area for that position. The different areas (labels for the classification) are:

Business
Technical
Sales-Marketing
Other

In the example, CEO and Founder would return Business.

There is an analogous project but it classifies according to the level of the position: Classification of job positions by level.

Implementation

This project is programed using the Python language. The trained classifiers are implemented in the Scikit Learn library, a set of tools for machine learning in Python. Two classification classes are studied:

Needed libraries

If you use pip and virtual environments, you can install easily the needed libraries, such as pandas and scikit learn:
pip install -r requirements.txt.

Process data

Since both classifiers belong to supervised learning, they are trained using manually classified data, that you can see on data_process/data_sets/classified_titles.tsv. That is a tab-separated-values file, that has two columns in the form:
<job position> | <classification for the job position>.
The script data_process/tsv_to_dataframe.py takes the tsv file and:

Generates a dataframe that represents the job positions and corresponding classifications.
Split the dataframe into train(X) and test(y) set.
Normalizes the dataframe according to a defined criteria.
Stores X_train, X_test, y_train, y_test sets.

Training and tuning

A classifier has:

parameters: values that corresponds to the mathematical model, that are adjusted after the training.
hyper-parameters: values related to the way of training, that are adjusted using a selected part of the training set.

It's possible to use a fit function to train and adjust the params. Besides, Scikit learn provides a tool named Pipeline, and GridSearchCV in order to make exhaustive search to achieve the hyperparams that optimize the results.

Script execution

The steps are the same as the classification by level:

Run data_process/tsv_file_to_dataframe.py to extract the data from the tsv file and split the dataset.
Run <clf name>_fit_tune_classifier.py to fit and tune the classifier. fit is to learn and fit the model to the train set. tune is to search for the optimal combination of the hyperparams, the ones that achieves better results (tuning may take a while).
Run <clf name>_test_classifier.py to test the trained classifiers and show the results. Besides, a classified example set is stored in test_data/<clf name>_results.tsv.

Notes:

<clf name> can be mlp or sgd, depending on the classifier.
mlp_fit_tune_classifier.py can take days to complete the tuning.

Results

These are the results of each classification class:

MLP

                  precision    recall  f1-score   support

       Business       0.91      0.92      0.91        63
          Other       0.79      0.86      0.83        44
Sales-Marketing       0.90      0.90      0.90        10
      Technical       0.86      0.73      0.79        33

       accuracy                           0.86       150
      macro avg       0.86      0.85      0.86       150
   weighted avg       0.86      0.86      0.86       150

SGD

                  precision    recall  f1-score   support

       Business       0.97      0.97      0.97        63
          Other       0.95      0.95      0.95        44
Sales-Marketing       0.89      0.80      0.84        10
      Technical       0.97      1.00      0.99        33

       accuracy                           0.96       150
      macro avg       0.95      0.93      0.94       150
   weighted avg       0.96      0.96      0.96       150

SGD has better average than MLP. SGD uses SVM model, a good model for text classification.