What is it about?
Data and code to make it easy to use entities on the MS MARCO collections for information retrieval research. MMEAD offers a JSON specification for sharing entity links and to simplify their usage. Out-of-the-box, entity links produced by the Radboud Entity Linker (REL) are provided on all MS MARCO collections, and Facebook's BLINK are provided for MS MARCO v1. We also provide the Wikipedia2Vec embeddings of the Wikipedia dump that we linked to.
Featured Image
Photo by Nathan Dumlao on Unsplash
Why is it important?
As shown in the paper, recognizing entities in the Web documents and queries can help improve retrieval effectiveness and facilitate new ways for users to navigate the document collection.
Perspectives
The ease of use of entity annotations that we provide with MMEAD aims to assist colleague IR researchers to study the role of explicit knowledge in achieving high retrieval effectiveness, and may even help lift analyses of retrieval experiments beyond the shallow (but oh so common) "NDCG improved by x% using model X".
Prof.dr.ir. Arjen P. de Vries
Radboud Universiteit
Read the Original
This page is a summary of: MMEAD: MS MARCO Entity Annotations and Disambiguations, July 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3539618.3591887.
You can read the full text:
Resources
Contributors
The following have contributed to this page
