The week's pick
Random Articles
Reseach Article
Optimizing Bidirectional LSTM Networks with Temperature-Scaled Sigmoid Activation for Enhanced Fake Profile Detection on Social Media
| International Journal of Computer Applications |
| Foundation of Computer Science (FCS), NY, USA |
| Volume 186 - Number 67 |
| Year of Publication: 2025 |
| Authors: Govind Singh Mahara, Sharad Gangele, Mangala Prasad Mishra |
10.5120/ijca2025924492
|
Govind Singh Mahara, Sharad Gangele, Mangala Prasad Mishra . Optimizing Bidirectional LSTM Networks with Temperature-Scaled Sigmoid Activation for Enhanced Fake Profile Detection on Social Media. International Journal of Computer Applications. 186, 67 ( Feb 2025), 53-65. DOI=10.5120/ijca2025924492
Abstract
This study presents an optimized Bidirectional Long Short-Term Memory (Bi-LSTM) network that integrates a temperature-scaled softmax function to improve the detection of fake profiles on Instagram and other social media platforms. By incorporating temperature scaling, the model effectively reduces overconfidence in its probability predictions, leading to more calibrated and reliable outputs. [1] Is explained by temperature scaling in single parameter variation, In this setup, the temperature-scaling function acts as a modifier on the softmax output layer, which enhances the model’s confidence alignment and mitigates issues that arise from predictions that might otherwise be excessively certain, particularly in ambiguous cases.The research explores the impact of varying learning rates and temperature values across multiple experimental setups. By fine-tuning these parameters, the model achieves an optimized balance in accuracy and stability, enhancing its overall performance. The Bi-LSTM model outperformed traditional methods in several key metrics, including accuracy, precision, recall, and F1 score. This suggests that the temperature-scaled Bi-LSTM approach is better suited for tasks that require nuanced classification in binary settings, such as differentiating between real and fake profiles. Additionally, the proposed framework’s flexibility allows it to be readily adapted to other binary classification problems on social media platforms, including spam filtering and fraud detection. This adaptability extends the potential applications of the model across various domains where binary classification is essential for safeguarding platform integrity.The experiments for this research were conducted on Google Colab, a popular cloud-based platform offering free access to GPUs, making it a suitable environment for deep learning projects
References
- Guo, C., Pleiss, G., Sun, Y, 2017. Weinberger, K. Q, On Calibration of Modern Neural Networks
- Ferrara, E. 2016.The Rise of Social Bots.
- Subrahmanian, V.S 2016. The DARPA Twitter Bot Challenge.
- Kumar, S., et al. 2018. False Information on Web and Social Media: A Survey. Proceedings of the VLDB Endowment.
- Cresci, S., et al. 2015.Fame for Sale: Efficient Detection of Fake Twitter Followers. Decision Support Systems.
- Yang, Singh and Menczer 2024.Characteristics and prevalence of fake social media profiles with AI-generated faces
- Varol, O., et al. 2017. Online Human-Bot Interactions: Detection, Estimation, and Characterization.
- Almaatouq, A., et al. 2016. If It Looks Like a Spam Bot, and Behaves Like a Spam Bot, It Must Be a Spam Bot: Analysis and Detection of Microblogging Spam
- Sergio A. Balanya, Juan M, Daniel R 2024.Adaptive temperature scaling for Robust calibration of deep neural networks
- Andy S,Dorsa S,Stefano E 2023. Long Horizon Temperature Scaling
- Graves, A,Schmidhuber J 2005.Framewise phoneme classification with bidirectional LSTM and other neural network architectures. Neural Networks
- Schuster, M.,Paliwal, K. K. 1997. Bidirectional recurrent neural networks
- Wang, Z., Yan, W., & Oates, T. 2017. Time series classification from scratch with deep neural networks: A strong baseline
- Kim, Y., Jernite, Y., Sontag, D., & Rush, A. M. 2016. Character-aware neural language models.
- Cui, Z., Chen, W.,Chen, Y. 2016.Multi-scale convolutional neural networks for time series classification
- Karim, F., Majumdar, S., Darabi, H., & Harford, S. 2018, LSTM fully convolutional networks for time series classification
- Lipton, Z. C., Berkowitz, J., & Elkan, C. 2015,A critical review of recurrent neural networks for sequence learning
- Hochreiter, S., and Schmidhuber, J. 1997. Long short-term memory. Neural Computation.
- Hashida, S., and Tamura, K. 2019. Multi-channel MHLF: LSTM-FCN using MACD-histogram with multi-channel input for time series classification.
- Mahara, G. S., and Gangele, S. 2021. A comprehensive survey of social network analysis-based anomaly detection techniques with soft computing.
- Ishlam, T. 2022. A proposed Bi-LSTM method for fake news detection. arXiv preprint.
- Mahara, G. S., and Gangele, S. 2022. A survey on detecting fake news in social media with AI: Challenges and possible directions.
- Mahara, G. S., and Gangele, S. 2022. Fake news detection: A RNN-LSTM, Bi-LSTM-based deep learning approach.
- Shu, K., Zhou, X., Wang, S., and Zafarani, R. 2019. The role of user profiles for fake news detection.
- Staudemeyer, R. C., and Morris, E. R. 2019. Understanding LSTM – a tutorial into Long Short-Term Memory recurrent neural networks.
- Cook, J., and Ramadas, V. 2020. When to consult precision-recall curves.
- Mao, A., Mohri, M., and Zhong, Y. 2023. Cross-entropy loss functions: Theoretical analysis and applications.
- Davis, J., and Goadrich, M. 2006. The relationship between precision-recall and ROC curves.
Index Terms
Keywords