What is it about?
When searching, we usually want to find documents, a list of entities or a direct answer, about a given topic or for a given question. Combining heterogeneous information sources is a challenge and, even more, cross-referencing information from such sources during retrieval tasks. With hypergraph-of-entity, we propose (1) a joint representation of terms, entities and their relations, from corpora and knowledge bases, in order to (2) solve multiple tasks of entity-oriented search, based on a universal ranking function.
Featured Image
Why is it important?
So far, we have been relying on the inverted indexes and triplestores, which are disjoint representations, to solve entity-oriented search tasks. Usually, ranking solutions revolve around translating the problem into the inverted index (i.e., somehow representing the entities and their relations as virtual documents), translating the problem into a triplestore (e.g., using the documents to build SPARQL queries), or using learning to rank to combine signals from both representations. Few data structures had been proposed to combine both data sources and, to our knowledge, no data structure had been proposed that would do so in a seamless manner, using the same units of representation (i.e., terms and entities as nodes, and hyperedges as documents).
Perspectives
The idea of building a unified framework for entity-oriented search, or even information retrieval in general, while a stranger in this area, has been explored in many other areas as the ultimate problem. From physics to machine learning, unified theories have been considered the holy grail (e.g., "The Master Algorithm"). With the hypergraph-of-entity, I personally wanted to launch this same challenge to the information retrieval community, because I know that cognitive search will only be possible by cross-referencing multiple leads, as opposed to using a more modular approach to retrieval. I am aware of the challenge this poses and the complexity of the problem, but I still believe it's a challenge worth pursuing.
José Devezas
INESC TEC
Read the Original
This page is a summary of: Hypergraph-of-entity, Open Computer Science, January 2019, De Gruyter,
DOI: 10.1515/comp-2019-0006.
You can read the full text:
Contributors
The following have contributed to this page
