| CPC G06F 16/3344 (2019.01) [G06F 16/31 (2019.01); G06F 16/3347 (2019.01); G06F 16/355 (2019.01); G06F 40/186 (2020.01); G06F 40/279 (2020.01); G06F 40/295 (2020.01); G06N 20/20 (2019.01); G06Q 50/18 (2013.01)] | 16 Claims |

|
1. A method of querying a computer database system, the method comprising:
determining, by a database system, a set of natural language questions associated with a given source of questions, the set of natural language questions comprising:
a current natural language question associated with the source; and
a set of prior natural language questions comprising natural language questions associated with the source and submitted prior to submission of the current natural language question;
determining, by a first question encoder of the database system, and based on the current natural language question, a first question model output comprising question vectors, the first question encoder employing a first bidirectional long-short-term memory (BiLSTM) model to generate a set of question vectors based on n-gram scores for n-grams in the current natural language question, an attention operation is conducted on the set of question vectors to generate an attention weighted set of question vectors, a concatenation operation is performed on the attention weighted set of question vectors to generate a concatenated set of question vectors, and the first question encoder employs a second BiLSTM model to generate, based on the concatenated set of question vectors, the question vectors;
determining, by a second question encoder of the database system and based on one or more questions of the set of prior natural language questions, a second question model output comprising context parameters;
determining, by a third question encoder of the database system and based on the question vectors and the context parameters, a set of concatenated question vectors for the question vector and the context parameters;
determining, by the database system, a selected database corresponding to the current natural language question, the selected database comprising features having features names and feature values;
determining, by a table encoder of the database system and based on the features of the selected database, a table encoder model output comprising word vectors;
determining, by decoder of the database system and based on the set of concatenated question vectors and the word vectors, a set of strings comprising a feature name of the feature names of the data table and a database language operator;
determining, by a transformer model of the database system and based on the set of strings and the database language operator, a set of values;
generating, by the database system based on the set of strings and the set of values, a database language query for accessing data stored in a database; and
executing, by the database system, the database language query to retrieve, from a datastore, a set of data corresponding to the database language query,
presenting the set of data retrieved to a user in response to the execution of the database language query.
|