Pre-extracted features for Lightweight MMHA multimodal sentiment analysis

This repository contains pre-extracted multimodal features (text, audio, and video) from the CMU-MOSI dataset, used in the paper "A Deep Learning Framework for Multimodal Sentiment Analysis Using Attention Mechanisms" for multimodal sentiment analysis with the Lightweight Multi-Modal Hybrid Attention (MMHA) framework. Text features were extracted using [text model, AoBERT], audio features using [audio tool/model, Whisper], and video features using [visual tool/model, ResNet-50]. File format: [e.g., .npy]. Features are provided for the train and validation. The original CMU-MOSI dataset is publicly available from its authors; this repository provides only the derived features.