Reverse Engineering the YouTube Algorithm

Built a graph crawler to map YouTube’s recommendation algorithm to configurable depth, constructing node-edge graphs across 10K+ videos and applying ForceAtlas2 for community detection. Designed a custom weighted scoring algorithm with modularity-based classification, achieving cluster-level ranking via in-degree sampling.

The work was published as a research paper presenting the parametric algorithmic transformer-based framework for weighted video analysis, modeling the relationships between recommended content at scale.