Map paper abstracts into interpretable research themes and surface nearby work through vector similarity.
- Portfolio decision: New
- Learning type: Unsupervised learning
- Core method: TF-IDF, K-Means, cosine similarity, and SVD
- Execution: Notebook executed successfully with outputs committed
The project demonstrates how patterns, similarities, communities, or anomalies can be discovered without using a prediction target during model fitting. Results are interpreted as exploratory signals rather than ground truth.
A deterministic domain-shaped text corpus is generated inside the notebook. Hidden generator topics are used only for post-hoc diagnostics.
The notebook documents its data contract, reproducible seed, quality checks, and any synthetic proxy labels used only for retrospective diagnostics.
Data validation
↓
Feature representation and scaling
↓
Unsupervised model comparison
↓
Internal metrics and stability checks
↓
Interpretation, visualization, and limitations
The notebook uses method-appropriate evidence such as silhouette score, Davies-Bouldin index, cluster stability, retrieval similarity, graph structure, anomaly-score diagnostics, and post-hoc synthetic checks. No demo result is presented as a production benchmark.
├── research_paper_topic_mapping.ipynb
└── README.md
From the repository root:
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r "Unsupervised Learning Projects/requirements.txt"
jupyter lab "Unsupervised Learning Projects/Research Paper Topic Mapping/research_paper_topic_mapping.ipynb"Run the notebook from top to bottom. Saved outputs are included for review.
- Python
- Pandas and NumPy
- Scikit-learn
- Matplotlib and Seaborn
- Jupyter
Discovered groups and anomaly flags do not establish identity, intent, causality, risk, or business impact. Production use requires domain review, representative data, privacy controls, monitoring, and decision-specific evaluation.
Tajamul Khan — Data Scientist and AI Engineer
