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

Vishesh Bhardwaj, Aman Yadav, Srikireddy Dhanunjay Reddy, Tharun Kumar Reddy Bollu · 2025

Read Cognitive Inference based Feature Pyramid Network for Sentimental Analysis using EEG Signals by Vishesh Bhardwaj, Aman Yadav, Srikireddy Dhanunjay Reddy, Tharun Kumar Reddy Bollu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-18346.txt.

Introduction

Vishesh Bhardwaj leads a four author team reading feelings straight from brain waves. Head sensors pick up faint electric patterns that track mood as it shifts, which could aid mental health checks and kinder software. Their model reads the ZuCo 2.0 brain data, recorded with 128 channels while people read text with feeling.

A pyramid network pulls out patterns at several scales at once, from brief local flickers to broad waves, and a memory unit then tracks how those patterns unfold in time. The pairing scored 6.88 percent above past methods on the main set. It also held up on two other standard mood data sets, DEAP and SEED, a sign the design travels well.

Catalog record

CREATORVishesh Bhardwaj, Aman Yadav, Srikireddy Dhanunjay Reddy, Tharun Kumar Reddy Bollu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-18346

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Cognitive Inference based Feature Pyramid Network for Sentimental Analysis using EEG Signals Vishesh Bhardwaj, Aman Yadav, Srikireddy Dhanunjay Reddy, Tharun Kumar Reddy Bollu 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. I Introduction EEG is a widely used non-invasive technique for monitoring brain activity, capturing the brain’s electrical signals in real-time. These signals reflect various cognitive processes and emotional states, making EEG signal analysis a powerful tool in neuroscience research, clinical diagnostics, and even consumer applications such as brain-computer interfaces [ 1 ] . Traditionally, EEG signal analysis has relied on conventional signal processing techniques such as Fourier transforms, wavelet transforms, and power spectral density analysis to extract meaningful patterns [ 2 ] . In addition, some advanced techniques are implemented to understand the underlying connectivity among channels using topological approaches [ 3 ] , [ 4 ] , [ 5 ] . However, these methods often require handcrafted feature extraction, which may fail to capture the full complexity and temporal dependencies within the data. With the growing availability of large EEG datasets and advances in deep learning, the field has shifted towards automated, data-driven approaches that can uncover latent patterns in EEG signals without extensive feature engineering [ 6 ] . One such example of a dataset is ZUCO 2.0, which has been used in EEG-based language and sentiment understanding. It offers structured annotations with high spatial and temporal resolution, making it suitable for multiscale feature extraction [ 12 ] . EEG data often consist of various types of noise from muscle movements, eye blinks, and environmental interference; therefore, the preprocessing step becomes crucial for further analysis. Previous studies have used various techniques for the removal of noise to enhance the quality of the EEG signal. Common methods include band-pass filtering, Independent Component Analysis(ICA), and wavelet decomposition. In addition, normalization techniques such as min-max scaling and z-score standardization have been used to ensure consistency between different EEG channels and subjects [ 7 ] . Sentiment analysis, the task of detecting and classifying emotional states from EEG data, plays a crucial role in understanding human emotions and has diverse applications in fields such as mental health monitoring, human-computer interaction, and affective computing. It involves interpreting complex patterns of brain activity to infer the underlying emotional states, which can range from happiness and sadness to stress or excitement. The process typically requires the extraction of relevant features from EEG signals, followed by classification into predefined emotional categories [ 8 ] . Traditional approaches have relied on methods such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), including long short-term memory (LSTM) networks, to capture spatial and temporal dependencies in the data [ 9 ] , [ 10 ] , [ 11 ] . However, these techniques have limitations. CNNs are effective in learning spatial features from EEG signals, but often do not fully capture the temporal dynamics of the data [ 9 ] . Meanwhile, RNN-based models like LSTMs, while adept at handling sequential dependencies, struggle with computational efficiency and issues such as vanishing gradients, which can hinder their performance on large datasets or long sequences [ 10 ] . Given these challenges, there is a growing need for advanced architectures capable of extracting hierarchical multiscale features from EEG data while efficiently modeling temporal dependencies. FPN have shown great potential in this regard. Zheng et al. [ 13 ] proposed a collaborative and multilevel feature selection network that takes advantage of the FPN-based architecture to enhance action recognition by extracting discriminative characteristics at multiple spatial scales and levels. This approach demonstrates the effectiveness of FPNs in learning robust hierarchical characteristics, which can be adapted for EEG-based applications. For temporal modeling, Ravanelli et al. [ 14 ] introduced Light Gated Recurrent Units (Li-GRUs), a streamlined variant of traditional GRUs optimized for speech recognition tasks. Their model significantly reduces computational overhead while preserving high accuracy, showcasing its potential for sequence modeling in real-time applications. These innovations underscore the adaptability of both FPN and GRU-based architectures in various domains and inspire their application in EEG-based sentiment analysis. Our framework introduces three key novel components that differentiate it from existing approaches: • Proposed a novel cognitive feature pyramid network framework, enabling multiscale feature extraction to capture both high-level abstractions and fine-grained emotional patterns in EEG signals. • Implementation of FPN for the extraction of high and low-level features from EEG sensor data, facilitating superior performance through improved implementation of diverse feature maps of EEG data. • Implementation of a Cognitive Sequence Inference Engine for EEG-based sentiment analysis, enabling efficient capture of temporal dynamics and effective handling of large-scale data streams to enhance overall predictive accuracy. The rest of this article is organized as follows. Section II delves into the related studies that were done. Section III defines the data set used for sentiment analysis with its parameters. In addition, it explains the functional details of each and every block in the framework. Section IV explains the results obtained from the proposed model and compares them with existing models. Section V summarizes the paper with a brief discussion of the proposed framework and its performance. II Related Studies Although many studies have been conducted using EEG datasets, only a few have focused on proper model-based modifications. Various modifications can be applied to the entire pipeline, as explained below. II-A Feature Extraction Techniques for EEG Datasets Feature extraction for EEG datasets traditionally relies on frequency-domain analysis or time-domain features, such as power spectral densities. However, newer approaches leverage more advanced techniques such as wavelet transforms, sparse PCA, spatio-temporal field, and deep learning models to extract more useful characteristics [ 15 ] . A key challenge remains in selecting the most discriminative features that can capture the emotional states represented in the EEG data. Fuzzy logic models have brought about a revolutionary change in the accuracy of the models, with some achieving up to 98% accuracy, compared to previous traditional models. This has recently emerged as a promising approach in signal processing. Unlike traditional models that rely on crisp values, fuzzy models allow for more flexible and nuanced decision-making, improving the classification of

Read the complete text