Data exploration using natural language with data sampling

1 . A method comprising:

obtaining, during a session, a natural language query associated with a dataset;

causing a first machine learning model to transform the natural language query to a structured data query;

causing a second machine learning model to determine an intent distribution associated with the structured data query based on a query sequence including a plurality of structured data queries and corresponding results, where the second machine learning model classifies the query sequence into an intent cluster and the intent distribution indicates a probability of the structured data query being associated with the intent cluster;

causing a third machine learning model to determine a sample of the dataset based on the structured data query, a state of the session including the query sequence, the corresponding results, and the intent distribution, where the sample is selected from a plurality of samples generated from the dataset using a plurality of distinct sampling algorithms that are used to generate different subset of the dataset and the sample includes a subset of data included in the dataset based on a first sampling algorithm of a plurality of distinct sampling algorithms, the sample selected by the third machine learning model to minimize a divergence of the intent distribution based on a divergence metric that quantifies the difference between the intent distribution as applied to the dataset and the intent distribution as preserved in the sample; and

causing a query processor to perform the structured data query on the sample of the dataset and generate a visualization based on a result of the structured data query.

2 . The method of claim 1 , wherein the method further comprises determining a confidence value for a result of the structured data query based on the sample.

3 . The method of claim 2 , wherein the confidence value indicates an estimate of a variance associated with the result based on the sample.

4 . The method of claim 1 , wherein the second machine learning model is trained using historical data indicating the plurality of structured data queries executed by the query processor using the plurality of samples of the dataset.

5 . The method of claim 1 , wherein at least one sample of the plurality of samples is generated using a combination of samples of the plurality of sampling algorithms.

6 . The method of claim 1 , wherein the method further comprises causing a fourth machine learning model to generate a set of recommended queries of the dataset based on the intent distribution.

7 . The method of claim 1 , wherein the second machine learning model determines the intent distribution associated with the structured data query by at least:

determining an implicit intent associated with a query sequence comprising the structured data query; and

providing the structured data query and the implicit intent of the query sequence as input to the second machine learning model;

wherein the third machine learning model selects the sample based on an output from the second machine learning model.

8 . A non-transitory computer-readable medium storing executable instructions embodied thereon, that, as a result of being executed by a processing device, cause the processing device to perform operations comprising:

obtaining, during a session including a query sequence, a natural language query;

generating a query for a database based on the natural language query;

determining an intent distribution associated with the query based on an output of a first machine learning model, the intent distribution indicating a set of probabilities corresponding to a set of intent clusters associated with the query sequence;

selecting a sample of the database based on the query, the query sequence, corresponding responses, and the intent distribution associated with the query, wherein the sample is selected from a plurality of samples of the database generated based on a plurality of distinct sampling strategies that are used to generate the plurality of samples and selected, by a second machine learning model, to minimize a divergence associated with the intent distribution; and

obtaining a result by at least executing the query on the sample.

9 . The medium of claim 8 , wherein the operations further comprise:

generating a set of recommended queries based at least in part on the query and the intent distribution;

obtaining a selection of a recommended query of the set of recommended queries; and

modifying the intent distribution based on the recommended query.

10 . The medium of claim 9 , wherein the set of recommended queries are generated based on a set of trees generated representing the query sequence, where the set of recommended queries include a number of trees similar to a subset of trees of the set of trees.

11 . The medium of claim 9 , wherein the set of recommended queries is a generated based on a history of queries associated with the database.

12 . The medium of claim 9 , wherein selecting the sample further comprises causing the second machine learning model to take an input a state associated with the session indicating a set of previous intent distributions associated with previous queries included in the query sequence.

13 . The medium of claim 12 , wherein the second machine learning model takes as an input a sequence of queries executed during a prior session.

14 . The medium of claim 8 , wherein the operations further comprise:

determining a confidence score associated with the query based on the sample; and

causing a user interface to display a representation of the confidence score.

15 . The medium of claim 8 , wherein the operations further comprise:

generating a visualization associated with the query based on the result; and

causing a user interface to display the visualization.

16 . A system comprising:

a memory component; and

a processing device coupled to the memory component, the processing device to perform operations comprising:

obtaining, through a user interface, a natural language query associated with a dataset and a session including a set of queries of with the natural language query is a member;

determining a query of the dataset based on the natural language query;

obtaining an intent distribution associated with the query from a first machine learning model taking as an input the query, the intent distribution indicates a set of probabilities corresponding to a set of intent clusters generated based on the set of queries;

determining a sample of the dataset from a plurality of samples of the dataset based on a first output of a second machine learning model, where the second machine learning model takes as an input the query, the intent distribution, and the set of queries, wherein the sample includes a subset of data included in the dataset selected based on a sampling strategy, the sample selected by a third machine learning model to minimize a divergence of an intent of the intent distribution based on a divergence metric that quantifies a difference between the intent as applied to the dataset and the intent as preserved in the sample;

generating a set of recommend queries based on a second output of the third machine learning model, where the third machine learning model takes as an input the intent; and

causing a user interface to display the set of recommended queries and a result by at least executing the query on the sample.

17 . The system of claim 16 , wherein the third machine learning model further takes as an input a state of the session.

18 . The system of claim 17 , wherein the generating the set of recommend queries further comprises:

modeling a set of previous executed queries as a set of trees, where nodes of the trees indicate executed queries and associated intents;

modeling the set of queries executed during the session as a second set of trees; and

identifying the set of recommend queries by at least comparing a distance of trees in the set of trees and the second set of trees where the intent matches the associated intents.

19 . The system of claim 17 , wherein the second machine learning model further generates a selection of the sample as the first output, such that the sample is selected to reduce a divergence associated with the intent as a result of executing the query on the sample.

20 . The system of claim 16 , wherein the operations further comprise determining a confidence associated with the sample, where the confidence indicates an estimate of a variance associated with the result based on the sample.