Al-Khwarizmi College of Engineering held a master defense about Popular Account Detection in Social Media Network based Deep Learning and Optimization Algorithms

On Wednesday, March, 4^th ,2026 Al-Khwarizmi College of engineering held a master thesis defense in the field of Communication and information engineering titled “Popular Account Detection in Social Media Network based on Deep Learning and Optimization Algorithms”, for the master candidate (Faten Muhammad Karim) at the thesis defense hall, where a degree of passing with distinction was awarded for her research findings.

The committee is composed of the following members:

• Assistant Professor Dr. Iman Salih Karim, Chair
• Assistant Professor Dr. Ahmed Sattar Hadi, Examiner
• Assistant Professor Dr. Ammar Adel Hasan, Examiner
• Lecturer Dr. Fatima Bahjit, Supervisor and committee member

This work addresses the problem by modeling influencer data as a dynamic sequence of monthly heterogeneous graphs and proposes a two-stage framework:

the first stage learns node embeddings using HGNN with BiGRU, while the second stage performs temporal influencer ranking using BiLSTM with a Multi-head Self-attention mechanism. The framework is evaluated using both chronological split and influencer-ID split protocols with ranking metrics such as NDCG and RBP. Experimental results show that the HGNN (pre-BiGRU) model achieves the best performance under chronological split (NDCG@200 = 83.6%), while the HGNN-only model achieves the highest-ranking quality under influencer-ID split (NDCG@200 = 88.4%). The results indicate that separating representation learning from the downstream ranking stage improves the effectiveness of influencer ranking systems.

The results were published in international scientific journals that indexed in Q2 &Q3, reflecting the scientific relevance and the practical significance in the field of social media networks data analysis. The study also recommended splitting node embodying learning from task specific learning as it contributes to reducing both time and memory complexity, furthermore, this would extend the proposed system to include other tasks such as content popularity on social media. As well as, modeling social network environment as a Heterogeneous Data Object, which provides high efficiency by reducing memory consumption and execution time.