Cognitive Inference based Feature Pyramid Network for Sentimental Analysis using EEG Signals

TL;DR

This paper introduces a novel EEG-based sentiment analysis model combining a Feature Pyramid Network with a GRU, achieving improved accuracy in classifying emotional states from EEG signals.

Contribution

It adapts Feature Pyramid Networks for multiscale EEG feature extraction and integrates temporal modeling, advancing sentiment analysis methods using EEG data.

Findings

Achieved a 6.88% performance improvement over existing methods.

Validated effectiveness on ZUCO 2.0, DEAP, and SEED datasets.

Enhanced multiscale feature extraction for EEG-based sentiment classification.

Abstract

Sentiment analysis using Electroencephalography (EEG) sensor signals provides a deeper behavioral understanding of a person's emotional state, offering insights into real-time mood fluctuations. This approach takes advantage of brain electrical activity, making it a promising tool for various applications, including mental health monitoring, affective computing, and personalised user experiences. An encoder-based model for EEG-to-sentiment analysis, utilizing the ZUCO 2.0 dataset and incorporating a Feature Pyramid Network (FPN), is proposed to enhance this process. FPNs are adapted here for EEG sensor data, enabling multiscale feature extraction to capture local and global sentiment-related patterns. The raw EEG sensor data from the ZUCO 2.0 dataset is pre-processed and passed through the FPN, which extracts hierarchical features. In addition, extracted features are passed to a Gated Recurrent Unit (GRU) to model temporal dependencies, thereby enhancing the accuracy of sentiment classification. The ZUCO 2.0 dataset is utilized for its clear and detailed representation in 128 channels, offering rich spatial and temporal resolution. The experimental metric results show that the proposed architecture achieves a 6.88\% performance gain compared to the existing methods. Furthermore, the proposed framework demonstrated its efficacy on the validation datasets DEAP and SEED.