Identifikasi Kepribadian Berbasis Teks dengan Fine-tuned Transformer Models

Hibatullah, Ulima Kaltsum Rizky (2025) Identifikasi Kepribadian Berbasis Teks dengan Fine-tuned Transformer Models. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211232-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Penelitian membahas identifikasi tipe kepribadian Myers–Briggs Type Indicator (MBTI) berbasis teks melalui pendekatan fine-tuned transformer model, dengan fokus pada perbandingan kinerja model FLAN-T5, T5, serta model shallow learning (K-Nearest Neighbors) dan deep learning tradisional (Multi-Layer Perceptron). Data yang digunakan berupa unggahan komentar pengguna berbahasa Inggris yang telah dilabeli tipe kepribadian MBTI. Proses analisis dilakukan melalui tahap pembersihan teks, lemmatisasi, tokenisasi, ekstraksi fitur menggunakan Term Frequency-Inverse Document Frequency (TF-IDF) dan Word2VEC, serta skenario augmentasi data untuk menangani ketidakseimbangan kelas. Hasil akhir berupa model berbasis Transformer memiliki kinerja tertinggi, dengan akurasi mencapai 76%. Hasil ini jauh melampaui model MLP dan K-NN, terutama ketika menggunakan representasi Word2VEC yang kinerjanya cenderung rendah. Analisis linguistik dengan WordCloud pada prediksi yang benar menunjukkan bahwa masing-masing tipe kepribadian memiliki pola bahasa yang distingtif, seperti perbedaan dalam ekspresi emosional, struktur kalimat, hingga kecenderungan penggunaan kata abstrak versus konkret. Penelitian ini menyimpulkan bahwa pendekatan model berbasis Transformer, mampu menangkap nuansa linguistik dalam teks dengan baik dan dapat diandalkan untuk prediksi kepribadian MBTI. Diharapkan, penelitian ini dapat menjadi dasar bagi pengembangan sistem analisis identifikasi kepribadian MBTI berbasis tulisan.

=====================================================================================================================================
This research discusses the identification of Myers–Briggs Type Indicator (MBTI) personality types based on text using a fine-tuned language model approach, focusing on the kinerjance comparison between FLAN-T5, T5, shallow learning models (K-Nearest Neighbors), and traditional deep learning models (Multi-Layer Perceptron). The data used consists of English-language user comment posts that have been labeled with MBTI personality types. The analysis process involves text cleaning, lemmatization, tokenization, feature extraction using Term Frequency-Inverse Document Frequency (TF-IDF) and Word2VEC, as well as data augmentation scenarios to address class imbalance. The final results show that Transformer-based models achieved the highest kinerjance, with accuracy reaching 76%. This significantly outperforms both MLP and K-NN models, especially those using Word2VEC feature representations, which tend to have lower kinerjance. Linguistic analysis using WordCloud on correctly predicted samples reveals that each personality type demonstrates distinctive language patterns, such as differences in emotional expression, sentence structure, and the tendency to use abstract versus concrete words. This study concludes that Transformer-based models are capable of capturing nuanced linguistic patterns in text effectively and can be relied upon for MBTI personality prediction. It is expected that this research can serve as a foundation for the development of MBTI personality identification systems based on written text.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Analisis Linguistik, Identifikasi MBTI, K-NN, MLP, Transformer K-NN, Linguistic Analysis, MBTI Identification, MLP, Transformer
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Ulima Kaltsum Rizky Hibatullah
Date Deposited:	30 Jul 2025 06:17
Last Modified:	30 Jul 2025 06:17
URI:	http://repository.its.ac.id/id/eprint/123199

Actions (login required)

View Item