IMFD participates in the creation of G-Core, a new language that optimizes searches
Every time a user performs a search, the data is stored on a server. The same happens with almost everything we share via e-mail, social media, or mobile networks. This information has reached such a volume that computer science devotes an entire line of research to the study of new models that can improve data storage systems, data relationships, and data comprehension.
Currently, large companies and services work with a model called knowledge graphs, a technology that has enabled the delivery of information related to a given search.
“Tourist attractions” associated with a city or the “name of an actor” in a movie are just a few examples. It is graph-based methods that allow Google to infer, for instance, that along with wanting to visit Park Güell and La Rambla in Barcelona, a tourist might also want to know opening hours and which restaurants are recommended. Or that, along with the forgotten name of an actor from a movie, they might like to know what other films that actor has appeared in, or which actresses and directors are related.
More and better searches
However, computer science aims to make these searches even smarter: “The goal is to use the information available in any database to answer complex questions that involve discovering the relationships between variables,” explains Pablo Barceló, Deputy Director of the Millennium Institute for Foundational Research on Data (IMFD) at Universidad de Chile.
“Knowledge graphs are a way of organizing the networks that exist between data, and are, so to speak, the architecture on which Google and other search engines run, constituting a new paradigm of information management used today by major technology companies,” Barceló adds.
Industry and academia working toward a better query language
The Millennium Institute for Foundational Research on Data has extensive national and international experience in the study of knowledge graphs, which is why it was invited to be part of an international team of 12 researchers who, under the LDBC Council and alongside companies such as IBM, Oracle, and Neo4J, have spent two years working on G-Core, a query language that makes it possible to discover, extract, and understand the most relevant relationships between pairs of data.
“A piece of data only gains value in relation to another,” explains Claudio Gutiérrez, researcher at the institute at Universidad de Chile: “Its richness lies not in the information itself, but in the connections that can be discovered or determined between one node and another,” he details.
The newly created query language was presented by the international team of which IMFD is part at the international conference Sigmod/PODS 2018, one of the most important global conferences on data management, recently held in Houston, United States.
“G-Core is the only language that discovers the paths between one piece of data and another, making it possible to generate extremely valuable information to reveal, for example, relationships of power, business, or behavior between one node and another,” notes Claudio Gutiérrez.
For this reason, the researchers foresee important applications in the field of social sciences, though it could be applied in any area. “We work on developing and improving methods to access information, which is why, from a scientific standpoint, this advance could have a major impact on knowledge graphs,” the researcher states.
The institute hopes that this language will soon translate into an application that can be used by professionals who need to extract information from complex, highly interrelated data sets.
Scientifically proven
Marcelo Arenas, Director of the Millennium Institute for Foundational Research on Data and academic at Pontificia Universidad Católica, explains that today the most widely used query language for graphs is Cypher, but that G-Core offers two specific improvements: “First, we mathematically proved that all queries made will return results, something Cypher is not able to guarantee,” he details.
Second, while Cypher’s answers are returned in table format, G-Core’s are returned as knowledge graphs, which means that, potentially, infinite queries can be made on top of a single query, further refining the searches,” he concludes.