Knowledge-Graph-Embedding Learning-to-Rank for Support Engineer Recommendation
An approach to routing customer-support incidents that treats engineer assignment as a learning-to-rank task and represents the support ecosystem as a knowledge graph. Embeddings learned from multiple heterogeneous sources—incident descriptions, affected components, engineer expertise, knowledge-base and response text, and historical 'swarm' collaborations—serve as ranking features, producing recommendations of the best engineer or swarm for an incident that outperform text-similarity baselines like TF-IDF.
2501.00461
This paper frames product-support triage as a learning-to-rank problem: given a customer incident, and possibly a set of engineers already working it as a 'swarm', produce a ranked list of the engine…