Wikipedia Entity Embeddings and Metadata for Entity-Aware Document Ranking (enwiki-20190701, 300-dim)

A comprehensive dataset for entity-aware document ranking and retrieval systems, consisting of two complementary files: 1. mmead_entities.wikipedia2vec.jsonl.gz (15.4 GB): Maps Wikipedia entities to their Wikipedia2Vec embeddings. Each entry contains entity_id, entity_name, wiki2vec_entity identifier, and a 300-dimensional embedding vector. Approximately 11.8 million entity embeddings from English Wikipedia (2019-07-01). 2. entity_metadata.jsonl.gz (683 MB): Contains entity metadata with id, title, and contents (Wikipedia text) for each entity. This provides the textual content that corresponds to the embeddings. These datasets support the QDER (Query-Specific Document and Entity Representations) and DREQ (Document Re-ranking Using Entity-Based Query Understanding) models for entity-oriented neural information retrieval. The embeddings enable query-relevant entity emphasis in document representations and entity-centric ranking approaches.