2024

Mentions graph — a trusted network of energy experts on Twitter

Graph traversal and active-node classification to surface a trusted community of energy experts from the open Twitter mentions/retweet network.

Built with NetworkX and the Louvain community-detection algorithm, the project treats a mentions-and-retweets graph as a noisy social substrate and uses active node classification to isolate a trusted community of domain experts in the energy sector. The pipeline combines graph traversal with transformer-based textual features to propagate labels to unseen nodes while keeping a human-in-the-loop for uncertain cases.

The work was published in the International Journal of Computer Science and Mobile Applications (2024) as Building a Trusted Network of Energy Experts on Twitter through Graph Traversal and Active Node Classification, with Eni’s Advanced Analytics team.

Under the hood the system relies on a hybrid of structural centrality, semantic similarity from fine-tuned transformers and iterative relabelling under weak supervision, yielding a directed subgraph that materially outperforms keyword-only baselines on expert recall.