Using transformers and Bi-LSTM with sentence embeddings for prediction of openness human personality trait
Understanding human personality traits is significant as it helps in decision making related to consumers’ behavior, career counselling, team building and top candidates’ selection for recruitment. Among various traits, openness is essential as it shows both diverse aspects of sensitive nature or in...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
PeerJ Inc.
2025-05-01
|
| Series: | PeerJ Computer Science |
| Subjects: | |
| Online Access: | https://peerj.com/articles/cs-2781.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Understanding human personality traits is significant as it helps in decision making related to consumers’ behavior, career counselling, team building and top candidates’ selection for recruitment. Among various traits, openness is essential as it shows both diverse aspects of sensitive nature or intuitive nature. The individuals having a sensing nature tends to be more practical and prefer to focus on concrete information whereas the users having intuitive trait type is characterized by a focus on abstract ideas, creative thinking and future-oriented perspectives. In this research work, we aim to explore diverse natural language processing (NLP) based features and apply state of the art deep learning algorithms for openness trait prediction. Using standard Myers-Briggs Type Indicator (MBTI) dataset, we propose the use of the latest deep features of sentence embeddings which captures contextual semantics of the content to be used with deep learning models. For comparison, we explore textual features of Frequency-Inverse Document (TF-IDF) and parts of speech (POS) tagging with machine learning models and deep features of word2vec and global vectors for word representation (GloVe) with deep learning models. The comprehensive empirical analysis reveals that TF-IDF used with gradient boosting achieves high accuracy of 90% whereas, the deep feature of sentence embeddings when used and with deep model bidirectional long short-term memory (Bi-LSTM) achieves 90.5% accuracy. The best results have been achieved using the latest Transformer-based DistilBERT, which achieves the highest accuracy of 92% outperforming the existing studies in relevant literature. |
|---|---|
| ISSN: | 2376-5992 |