[Paper Review] Personality Type Based on Myers-Briggs Type Indicator with Text Posting Style by using Traditional and Deep Learning

Sakdipat Ontoum, Jonathan H. Chan|arXiv (Cornell University)|Jan 21, 2022

Customer churn and segmentationBusiness, Management and Accounting19 citations

TL;DR

This study proposes a machine learning framework to predict Myers-Briggs Type Indicator (MBTI) personality types from text posts using traditional (Naive Bayes, SVM) and deep learning (Recurrent Neural Networks) models. The RNN-based approach, enhanced with word embeddings and bidirectional LSTM, achieved the highest accuracy, outperforming other models across all four MBTI dimensions, with an F1-score mean of approximately 76%.

ABSTRACT

The term personality may be expressed in terms of the individual differences in characteristics pattern of thinking, feeling, and behavior. This work presents several machine learning techniques including Naive Bayes, Support Vector Machines, and Recurrent Neural Networks to predict people personality from text based on Myers-Briggs Type Indicator (MBTI). Furthermore, this project applies CRISP-DM, which stands for Cross-Industry Standard Process for Data Mining, to guide the learning process. Since, CRISP-DM is kind of iterative development, we have adopted it with agile methodology, which is a rapid iterative software development method, in order to reduce the development cycle to be minimal.

Motivation & Objective

• To develop a machine learning-based system for predicting MBTI personality types from social media text posts.
• To compare the performance of traditional machine learning models (Naive Bayes, SVM) with deep learning (RNN) on MBTI classification.
• To apply the CRISP-DM framework with agile methodology to streamline the iterative development of the personality prediction model.
• To evaluate model performance across all four MBTI dimensions (e.g., Perception vs. Judgment) and identify classification weaknesses.
• To explore the feasibility of using forum-based text data (e.g., Personality Cafe) as a reliable source for personality prediction.

Proposed method

• Employed the CRISP-DM framework with agile methodology for iterative data mining and model development.
• Preprocessed text data by removing URLs, special characters, stop words, and performing case normalization and stemming.
• Used Scikit-learn to implement Naive Bayes (multinomial-NB) and Support Vector Machines (SVM) with linear kernel and regularization.
• Built a deep learning model using TensorFlow with word embeddings (vocabulary size 256), CONV1D layers for sequential feature extraction, and bidirectional LSTM (64 units) for bidirectional context learning.
• Split the dataset into 75% training and 25% testing using Scikit-learn’s train_test_split function.
• Evaluated models using confusion matrices, classification reports, and F1-score metrics to assess precision, recall, and F1 across all MBTI types.

Experimental results

Research questions

• RQ1 Which machine learning model—Naive Bayes, SVM, or Recurrent Neural Network—performs best in predicting MBTI personality types from text?
• RQ2 How does the performance of the models vary across the four MBTI dimensions (e.g., Perception vs. Judgment)?
• RQ3 To what extent does the use of bidirectional LSTM and word embeddings improve classification accuracy compared to traditional models?
• RQ4 What are the key limitations in predicting certain MBTI types, and how do confusion matrices reveal misclassification patterns?
• RQ5 Can social media text from forums like Personality Cafe serve as a valid data source for reliable personality prediction using ML?

Key findings

• The Recurrent Neural Network (RNN) model achieved the highest overall accuracy across all four MBTI personality types, outperforming both Naive Bayes and SVM.
• The F1-score mean for the RNN model was approximately 76%, indicating a balanced performance in precision and recall across all classes.
• The confusion matrix for the RNN model showed that most predictions were classified as true positives, with relatively few false positives, indicating strong classification consistency.
• Naive Bayes and SVM models exhibited significantly lower accuracy for the Perception (P) and Judgment (J) dimensions compared to the RNN model.
• The classification reports (Fig. 13 and 14) confirmed that the RNN model maintained higher precision and recall across all MBTI types, especially for the more challenging dimensions.
• Despite overall strong performance, the models showed general weaknesses in accurately classifying all four MBTI dimensions, suggesting inherent ambiguity or data imbalance in the dataset.