Zaina Ahmad Ali Alomari

Data Scientist Intern candidate with experience in Python, Pandas, Matplotlib, and model evaluation for language data. Analyzed BERT and GPT-4 retrieval behavior and delivered an Arabic sentiment analysis system with 86.35% accuracy. Turns unstructured text into measurable findings through preprocessing, comparison, and evaluation.

• Bidirectional Encoders vs. Generative Decoders: A Comparative Analysis of BERT and GPT-4 in Information Retrieval Tasks
• Analyzed architectural differences, attention mechanisms, and pre-training objectives of BERT and GPT-4
• Examined dense retrieval (DPR, ColBERT) and RAG pipelines
• Concluded complementary roles: BERT for scalable retrieval, GPT-4 for reasoning and generation
• Stop-Word Preservation in Modern Information Retrieval and Language Models
• Synthesized evidence showing stop-word removal degrades BERT-based retrieval accuracy
• Cited empirical studies: DPR top-1 accuracy drops from 45.9% to 37.8% when stop-words are removed
• Do Users Prefer Navigation Over Searching or Browsing? – A Review of Empirical Evidence
• Reviewed 8 peer-reviewed studies (2008-2024) on user retrieval behavior in PIM, web, and e-commerce contexts
• Synthesized findings: 56-95% of file retrievals use navigation; search is a "last resort" due to higher cognitive load

Passionate about digital art and designing.