AMPEL Framework: A Robust Multimodal Approach For Personality Prediction Using Automated Dataset Construction And Advanced Feature Learning

Abstract

The article presents the AMPEL Framework ( Automated Multimodal Personality Extraction and Learning ) to predict a personality based on multimodal data. The framework overcomes the problems of large dataset building and efficient feature learning, by combining audio, visual and textual features with real-life video entries. In the first step, audio integration, speech translators, feature extraction are performed to create a structured multimodal data, and pseudo-labels are obtained by behavioral hints. In the second phase, deep feature extraction, advanced feature engineering and hybrid feature selection methods are utilized to improve representation of features and minimizing redundancy. Further, dimensionality reduction techniques are used to streamline the feature space. The framework contains several machine learning models to predict personality, which make it flexible and strong. The proposed solution is a scale, multimodal personality, interpretable approach that is comprehensive and can be used in real-world scenarios.