Salesforce has made a breakthrough in the process of communicating with databases that require Structured Query Language (SQL) using a neural network.
This progress was achieved by researchers within the organisation’s AI team, having developed a system capable of handling SQL queries with natural language. A translation of the natural language is made by the technology, making it applicable to the queries.
Database querying is a complex process that limits the access to data to those that do not have the specific skills; this Salesforce project is intended to erase this barrier.
Google has also recently displayed interest in the potential that neural networks have to be disruptive, having acquired the startup, AIMatter. This company is rooted in image processing, and it is the creator of a popular app called Fabby.
Neural networks may also hold the key to enhancing the functionality of automation, an area of technology expected to be crucial in the future of spaces including cybersecurity.
The Salesforce AI research team detailed its findings in an academic paper; in conclusion it said: “We proposed Seq2SQL, a deep neural network for translating questions to SQL queries. Our model leverages the structure of SQL queries to reduce the output space of the model. As a part of Seq2SQL, we applied in-the-loop query execution to learn a policy for generating the conditions of the SQLquery, which is unordered in nature and unsuitable for optimization via cross entropy loss.”
Another natural language endeavour conducted by Salesforce involved the launch of a dataset collated from Wikipedia tables called WikiSQP, it is open-source, and holds in excess of 87,000 questions.
“We also introduced WikiSQL, a dataset of questions and SQL queries that is an order of magnitude largerthan comparable datasets. Finally, we showed that Seq2SQL outperforms attentional sequence tosequence models on WikiSQL, improving execution accuracy from 35.9% to 60.3% and logical formaccuracy from 23.4% to 49.2%,” said the Salesforce research paper.