Generating a response to a natural language command based on a concatenated graph

1. An apparatus comprising:

an input device;

a processor;

a memory that stores code executable by the processor to:

identify two or more relevant data sets in response to a natural language command received from the input device, wherein a first relevant data set comprises a first subject of the natural language command and a second relevant data set comprises a second subject of the first relevant data set;

generate a graph for each of the two or more relevant data sets;

concatenate the graphs into a concatenated graph by merging the second subject of a first graph for the first relevant data set and the second subject of a second graph for the second relevant data set;

in response to the concatenated graph exceeding a complexity threshold, determine that user feedback is required;

display the concatenated graph;

receive edits to the graph;

determine one or more solution sets from the edited concatenated graph;

calculate a solution score for each solution set a sum of a weighted number of subjects in the solution set divided by a sum of a weighted number of subjects in the natural language command;

modify each solution score as the solution score divided by a square root of a solution cost for the solution score;

select a solution set with a highest modified solution score for the response; and

generate a response to the natural language command based on the selected solution set of the concatenated graph.

2. The apparatus of claim 1 , wherein each graph comprises two or more of the subjects and one or more relationships between the two or more subjects.

3. The apparatus of claim 2 , wherein the code is further executable by the processor to prune irrelevant subjects.

4. A method comprising:

identifying, by use of a processor, two or more relevant data sets in response to a natural language command, wherein a first relevant data set comprises a first subject of the natural language command and a second relevant data set comprises a second subject of the first relevant data set;

generating a graph for each of the two or more relevant data sets;

concatenating the graphs into a concatenated graph by merging the second subject of a first graph for the first relevant data set and the second subject of a second graph for the second relevant data set;

in response to the concatenated graph exceeding a complexity threshold, determining that user feedback is required;

displaying the concatenated graph;

receiving edits to the graph;

determining one or more solution sets from the edited concatenated graph;

calculating a solution score for each solution set a sum of a weighted number of subjects in the solution set divided by a sum of a weighted number of subjects in the natural language command;

modifying each solution score as the solution score divided by a square root of a solution cost for the solution score;

selecting a solution set with a highest modified solution score for the response; and

generating a response to the natural language command based on the selected solution set of the concatenated graph.

5. The method of claim 4 , wherein each graph comprises two or more of the subjects and one or more relationships between the two or more subjects.

6. The method of claim 5 , the method further comprising pruning irrelevant subjects.

7. A program product comprising a non-transitory computer readable storage medium that stores code executable by a processor to perform:

identifying two or more relevant data sets in response to a natural language command, wherein a first relevant data set comprises a first subject of the natural language command and a second relevant data set comprises a second subject of the first relevant data set;

generating a graph for each of the two or more relevant data sets;

concatenating the graphs into a concatenated graph by merging the second subject of a first graph for the first relevant data set and the second subject of a second graph for the second relevant data set;

in response to the concatenated graph exceeding a complexity threshold, determining that user feedback is required;

displaying the concatenated graph;

receiving edits to the graph;

determining one or more solution sets from the edited concatenated graph;

calculate a solution score for each solution set a sum of a weighted number of subjects in the solution set divided by a sum of a weighted number of subjects in the natural language command;

modify each solution score as the solution score divided by a square root of a solution cost for the solution score;

select a solution set with a highest modified solution score for the response; and

generating a response to the natural language command based on the selected solution set of the concatenated graph.

8. The program product of claim 7 , wherein each graph comprises two or more of the subjects and one or more relationships between the two or more subjects.

9. The program product of claim 8 , the processor further pruning irrelevant subjects.