Machine learning to generate and evaluate visualizations

1. A method performed by one or more computers, the method comprising:

determining, by the one or more computers, properties of a dataset;

generating, by the one or more computers, visualization specifications that each define a different visualization for the dataset, wherein the visualization specifications specify different subsets of the dataset being illustrated with different visualization formats;

evaluating, by the one or more computers, the visualization specifications using a machine learning model trained based on user feedback for visualizations for multiple datasets;

selecting, by the one or more computers, a subset of the visualization specifications based on output of the machine learning model; and

providing, by the one or more computers, visualization data for display, the visualization data corresponding to the selected subset of the visualization specifications selected based on the output of the machine learning model.

2. The method of claim 1 , wherein determining properties of the dataset comprises determining structural properties of the dataset including at least one of a number of columns, a number of attributes, a number of metrics, a value for an aggregation for a data range, a span of values for a data range, a data type for a data range, or a statistical measure for a data range.

3. The method of claim 1 , wherein determining properties of the dataset comprises determining semantic properties of the dataset.

4. The method of claim 1 , wherein the different visualization formats comprise different visualization types including two or more from the group consisting of a graph, a line chart, a bar chart, a pie chart, a scatterplot, a heat map, a geographical map, a word-size map, a bubble chart, a hierarchy chart, a waterfall chart, a radar chart, a statistical chart, and an area chart.

5. The method of claim 1 , wherein the trained machine learning model has been trained to output scores indicating a level of usefulness of visualizations, where the scores are conditioned on properties of the datasets the visualizations are derived from, wherein the machine learning model is configured to generate the scores based on input indicating visualization properties and dataset properties, wherein the trained machine learning model has been trained based on user feedback data indicating actions of multiple users after being presented visualizations representing portions of different datasets.

6. The method of claim 5 , wherein the user feedback data comprises survey responses indicating ratings that users applied to visualizations.

7. The method of claim 5 , wherein the user feedback data comprises usage data comprising measures of user behavior, other than rating of visualizations, in response to the presentation of the visualizations.

8. The method of claim 7 , wherein the usage data comprises at least one of:

a number of times a visualization was viewed, interacted with, shared, or included in a document by one or more users; or

an amount of time that a visualization was viewed or interacted with by one or more users.

9. The method of claim 1 , wherein evaluating the visualization specifications comprises:

for each of the visualization specifications:

providing, to the trained machine learning model, input that indicates (i) properties of a visualization and (ii) properties of the dataset; and

receiving, as output of the trained machine learning model, a score for the visualization;

wherein selecting a subset of the visualization specifications comprises:

ranking the visualization specifications based on the scores provided as output of the machine learning model; and

selecting a highest-ranking subset of the visualization specifications.

10. The method of claim 1 , comprising receiving data indicating interaction with a control on a user interface; and

wherein providing the visualization data is performed in response to receiving the data indicating the interaction with the control on the user interface.

11. The method of claim 10 , further comprising determining a portion of the data set that has been selected using the user interface;

wherein selecting the subset of the visualization specifications comprises filtering the visualization specifications to exclude one or more visualization specifications that do not represent the determined portion of the data set.

12. The method of claim 11 , wherein determining that a portion of the data set has been selected comprises determining that a column of the data set has been selected; and

wherein selecting the subset of the visualization specifications comprises selecting only visualization specifications that are generated from or provide information regarding the selected column.

13. The method of claim 1 , wherein the machine learning model comprises an artificial neural network, a classifier, a support vector machine, a decision tree, a regression model, a clustering model, a Gaussian process model, a genetic algorithm, or a reinforcement learning model.

14. The method of claim 1 , further comprising identifying a user of a computing device;

wherein selecting the subset of the visualization specifications is personalized based on the identity of the user of the computing device; and

wherein providing the visualization data comprises providing the visualization data to the computing device for display by the computing device.

15. The method of claim 14 , further comprising identifying one or more topics based on a role of the user, a user profile for the user, or a usage history for the user;

wherein selecting the subset of the visualization specifications is personalized by adjusting scores for the visualization specifications to indicate increased relevance for visualization specifications associated with the one or more topics.

16. The method of claim 14 , further comprising determining, based on a usage history for the user, one or more visualization formats with which the user previously interacted;

wherein selecting the subset of the visualization specifications is personalized by adjusting scores to indicate increased relevance for visualization specifications specifying visualization formats that have at least a minimum level of similarity to the one or more visualization formats with which the user previously interacted.

17. The method of claim 10 , wherein evaluating the visualization specifications using the machine learning model is performed before receiving the data indicating the interaction with the control.

18. A system comprising:

one or more computers; and

one or more computer-readable media storing instructions that, when executed by the one or more computers, cause the one or more computers to perform operations comprising:

determining, by the one or more computers, properties of a dataset;

generating, by the one or more computers, visualization specifications that each define a different visualization for the dataset, wherein the visualization specifications specify different subsets of the dataset being illustrated with different visualization formats;

evaluating, by the one or more computers, the visualization specifications using a machine learning model trained based on user feedback for visualizations for multiple datasets;

selecting, by the one or more computers, a subset of the visualization specifications based on output of the machine learning model; and

providing, by the one or more computers, visualization data for display, the visualization data corresponding to the selected subset of the visualization specifications selected based on the output of the machine learning model.

19. The system of claim 18 , wherein the trained machine learning model has been trained to output scores for visualizations, wherein the trained machine learning model is configured to generate a score for a visualization with respect to a dataset based on input indicating (i) properties of the visualization and (ii) properties of the dataset.

20. The system of claim 18 , wherein the trained machine learning model has been trained based on examples of actions that users performed after being presented one or more visualizations, wherein at least some of the examples respectively indicate actions performed after presentation of visualizations for different data sets.

21. One or more non-transitory computer-readable media storing instructions that, when executed by the one or more computers, cause the one or more computers to perform operations comprising:

determining, by the one or more computers, properties of a dataset;

generating, by the one or more computers, visualization specifications that each define a different visualization for the dataset, wherein the visualization specifications specify different subsets of the dataset being illustrated with different visualization formats;

evaluating, by the one or more computers, the visualization specifications using a machine learning model trained based on user feedback for visualizations for multiple datasets;

selecting, by the one or more computers, a subset of the visualization specifications based on output of the machine learning model; and

providing, by the one or more computers, visualization data for display, the visualization data corresponding to the selected subset of the visualization specifications selected based on the output of the machine learning model.