Use setup.py to generate the configuration file, config.yml. After training the Word2Vec model, you need to run config_boundaries.sh to generate boundary values for the normalization process. This updates the config.yml with accordance to these values and it is crucial for the following steps.
In order to get Word2Vec representations properly, you need to train the Word2Vec library using train_model.sh. This step uses the Wikipedia archive for Turkish language to create a vector space to represent words.
You need to acquire OCEAN scores for your ground truth users. You may do it with a questionnaire, or other methods. We do not provide a concrete code block for this step, since it may vary vastly.
You also need to create the vector representation of your ground truth users. While determining the best pipeline via TPOT, you will need 5 different ground truth vectors with the following structure:
OCEAN_SCORE,d0,d1,d2,....,dn
The OCEAN_SCORE refers to the trait that the vector belongs to, and the rest is the values of the dimensions.
Place your ground truth vectors in the predictors/tpot folder with the following naming style: resultvector-{dimensions}-{top_words}-{window_size}-{trait}.csv where dimensions refers to the Word2Vec dimensions, top_words refers to the TFxIDF top word count, window_size refers to Word2Vec window size, and trait refers to the OCEAN trait.
After acquiring the ground truth vectors, you need to train your machine learning model to use with the system. In order to achieve the optimal training, we use TPOT. TPOT determines the optimal pipeline for each personality trait and reports the pipeline, which you should implement in your system.
Run TPOT via the teapot.py file inside the predictors/tpot folder. It should take a while depending on your system to determine the best prediction algorithm. For each personality trait, you must copy the pipeline and paste it inside the main.py to allow the system to predict personality types.