Improving Personality Detection With Hybrid 1d-Cnn and Lstm Models: An Empirical Study On Dataset Expansion

This study presents a hybrid deep learning model that combines 1D Convolutional Neural Networks (1D-CNN) and Long Short-Term Memory (LSTM) networks to improve personality trait prediction from text. Utilizing pre-trained FastText embeddings, the model captures both local and sequential linguistic features. Two datasets - the Essays dataset and the MBTI dataset - were used in both isolated and combined forms to evaluate the impact of data expansion. Extensive experiments across varying model depths (1, 3, and 5 layers) show that the hybrid architecture, particularly with three layers, consistently outperforms standalone CNN and LSTM models. The results demonstrate that combining architectural strengths and expanding data sources significantly enhances classification performance, especially for traits like Openness and Agreeableness. This work contributes a robust and scalable framework for personality detection and lays the groundwork for future research involving attention mechanisms or multimodal features.