Work
AI Research & Deep Learning
Multimodal Named Entity Recognition
Deep Learning Cross-Modal Feature Fusion for Social Media Entity Discovery
- By
- Muhammad Haider Akhtar
- Field
- AI Research & NLP
- Time
- Deep Learning Research
- Kind
- Founder's Professional Experience
The problem
Informal social media text contains short, slang-heavy, and unpunctuated sentences where textual-only Named Entity Recognition (NER) fails to identify persons, locations, and organizations accurately.
The work
Designed and evaluated a multimodal deep learning architecture in PyTorch combining BERTweet contextual language representations with CLIP visual embeddings through cross-modal attention mechanisms.
- PyTorch
- BERTweet
- CLIP Vision
- Transformers
- Cross-Modal Attention
- CUDA
What it does
- Verified research outcome from supplied CV: Achieved 78.6% overall F1 score across all entity classes
- Verified research outcome: Achieved exceptional 89.4% F1 score specifically on Person entity extraction
- Conducted extensive ablation studies validating that visual context significantly reduces false positives in ambiguous captions
Verified academic AI research project by founder Muhammad Haider Akhtar. Scores reflect documented research benchmarks.
NextBorehole Log ML AnalyzerApplied AI & Data Analytics