Personalized Query Expansion with Contextual Word Embeddings

Personalized Query Expansion, the task of expanding queries with additional terms extracted from user-related vocabulary, is a well-known solution to improve retrieval performance system w.r.t. short queries. Recent approaches rely on word embeddings select expansion texts. Although delivering promising results former embedding techniques, we argue that these methods are not suited for contextual embeddings, which produce unique vector representation each term occurrence. In this article, propose Expansion method designed solve issues arising use current based embeddings. Specifically, employ clustering-based procedure identify better represent user interests and diversity those selected expansion, achieving improvements up 4% best-performing baseline in MAP@100. Moreover, our approach outperforms previous ones efficiency, allowing us achieve sub-millisecond times even data-rich scenarios. Finally, introduce novel metric evaluate empirically show unsuitability when employed along cause selection semantically overlapping terms.