smesa / SentencePersonality

SentencePersonality computes personality traits, as described in Big5 model, from myPersonality dataset.

SentencePersonality

This package computes personality traits, as described in Big5 model, from myPersonality dataset.

The pipeline is the following:

• sentence encoding
• personality traits regression

The logical steps to reproduce the experiment

1. encode_and_map_sentences.py
   • input:
     • pretrained_bert/multi_cased_L-12_H-768_A-12
     • spacymoji vocabulary
     • myPersonalitySmall/statuses_unicode.txt
     • myPersonalitySmall/big5labels.txt
   • output:
     • train_whole_lines.csv
     • lines_skipped.csv
2. mse.py
   • input:
     • cls_table.csv
   • output:
     • predictions.txt
     • mse.txt
3. distributions.py
   • input:
     • predictions.py
     • predictions_eng.py //this file is used for comparison, change bert model model in previous steps
     • lines_skipped.py
     • train_whole_lines.csv
   • output:
     • data distributions as images
     • kullback leibler divergences between multilinglual model and english model

If you want to obtain Carducci et al. results take a look here:

* https://github.com/D2KLab/twitpersonality

If you want to reproduce scores by IBM Personality Insights:

* ibm_insights_script.py
* format the output to compare data distributions