Work

Doing sentiment analysis task using BERT NasNet-mobile optimized by improved multi-verse optimizer

Year: 2025

Type: article

Abstract: Identifying spam is essential to ensure the reliability of online platforms, such as Twitter and Reddit, which spammers can exploit to spread harmful content. This research presents an enhanced spam detection model that integrates NasNet-Mobile with the Improved Multi-Verse Optimizer (IMVO). The network has been strengthened by thorough preprocessing methods, such as tokenization, stemming, and stopword removal, to improve data quality and enhance the accuracy of classification. The suggested model has been assessed using the Reddit Spam Dataset and the Twitter Spam Dataset, focusing on metrics such as accuracy, precision, recall, and F1-score. The findings indicate that the NasNet-Mobile/IMVO + Pre model surpasses other traditional methods, including GRU, BiLSTM, and CNN, with an accuracy value of 97.95%, a precision value of 97.67%, a recall value of 97.82%, and an F1-score value of 97.74%. The metrics emphasize the importance of preprocessing steps and optimization in accomplishing efficient detection of spam. By balancing global and local search, IMVO helps the model avoid local optima and improves generalization. The results highlight the significance of combining a state-of-the-art optimization algorithm with a neural network model to effectively overcome real-world spam issues. The main contributions of this work are combining IMVO, a hybrid approach that uniquely considers the pitfalls of conventional spam detection. The contributions are (1) a novel preprocessing pipeline involving tokenization, stemming, and stopword removal to increase data quality; (2) the first application of IMVO for optimizing NasNet-Mobile architecture to rectify non-optimalities in local optima and to generalize across datasets; and (3) an empirical validation on Twitter and Reddit datasets establishing a state-of-the-art result against GRU, BiLSTM, and CNN baselines. This work advances spam detection by applying high-level optimization to deep learning models for an effective application in the real world.

Source: AIP Advances

DOI: 10.1063/5.0296192

Authors: Qian Zhang, Xiaochuan Guo, Lixia Liu, Mohammad Sarabi

Institutions: South China Normal University, Guangzhou College of Technology and Business, Sharif University of Technology, Islamic University of Najaf

Language: English

Topic: Spam and Phishing Detection

Subfield: Information Systems

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $1,350

Funders: Guangdong Provincial Pearl River Talents Program, Guangdong Academy of Sciences