Answering questions with artificial intelligence using tabular data

1. A computer-implemented method to answer a question using a data table, comprising:

receiving, by said computer, a user question and a target table containing a target cell corresponding to a target answer for said user question, said target cell corresponding to a target column and a target row;

generating, by said computer, a first classifier adapted to provide column correlation values reflecting the probability that a given column is said target column;

generating, by said computer, a second classifier adapted to provide row correlation values reflecting the probability that a given row is said target row;

applying, by said computer, said first classifier to the columns in the target table to determine a column correlation value for each column;

applying, by said computer, said second classifier to the rows in the target table to determine a row correlation value for each row;

suggesting, by said computer, as the target cell, a cell having elevated column and row correlation values relative to other cells in the target table;

receiving, by said computer, a set of training data; and

fine-tuning, by said computer using said training data, said first classifier and said second classifier.

2. The method of claim 1 , wherein said method further comprises:

the receiving, by said computer, of the set of training data including a training question, a training table, and identification of a ground truth target answer cell location within said training table, said target cell location corresponding to a target column and a target row;

the fine-tuning, by said computer, of said first classifier to provide, by said computer, column correlation values using said training data; and

the fine-tuning, by said computer, of said second classifier to provide, by said computer, row correlation values using said training data.

3. The method of claim 2 , wherein said fine tuning is conducted, by said computer, at least on part with weak supervision 3{3,10,17} of {2,9,16}.

4. The method of claim 1 , wherein said target cell has the highest row relevance value within a group of cells located in a column having a highest column correlation value.

5. The method of claim 1 , wherein said classifier is a pre-trained transformer-model adapted to execute a next sentence prediction task.

6. The method of claim 1 , wherein said classifier is a linear regression model adapted to classify feature vectors.

7. The method of claim 6 , further receiving, by said computer, a feature vector representation of said user question, said target table columns, and said target table rows; and

wherein said column correlation values and row correlation values are generated, at least in part, by said computer, by classifying said feature vector representation.

8. A system to answer a question using a data table, which comprises:

a computer system comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to:

receive a user question and a target table containing a target cell corresponding to a target answer for said user question, said target cell corresponding to a target column and a target row;

generate a first classifier adapted to provide column correlation values reflecting the probability that a given column is said target column;

generate a second classifier adapted to provide row correlation values reflecting the probability that a given row is said target row;

apply said first classifier to the columns in the target table to determine a column correlation value for each column;

apply said second classifier to the rows in the target table to determine a row correlation value for each row;

suggest as the target cell, a cell having elevated column and row correlation values relative to other cells in the target table;

receive, by said computer, a set of training data; and

fine-tune, by said computer using said training data, said first classifier and said second classifier.

9. The system of claim 8 , wherein said method further comprises:

the receiving of the set of training data including a training question, a training table, and identification of a ground truth target answer cell location within said training table, said target cell location corresponding to a target column and a target row;

the fine-tune of said first classifier to provide column correlation values using said training data; and

the fine-tune of said second classifier to provide row correlation values using said training data.

10. The system of claim 8 , wherein DEP of 2, wherein said fine tuning is conducted, at least on part with weak supervision 3{3,10,17} of {2,9,16}.

11. The system of claim 8 , wherein said target cell has the highest row relevance value within a group of cells located in a column having a highest column correlation value.

12. The system of claim 8 , wherein said classifier is a pre-trained transformer-model adapted to execute a next sentence prediction task.

13. The system of claim 8 , wherein said classifier is a linear regression model adapted to classify feature vectors.

14. The system of claim 13 , further receiving a feature vector representation of said user question, said target table columns, and said target table rows; and

wherein said column correlation values and row correlation values are generated, at least in part, by said computer, by classifying said feature vector representation.

15. A computer program product to answer a question using a data table, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to:

receive, using said computer, a user question and a target table containing a target cell corresponding to a target answer for said user question, said target cell corresponding to a target column and a target row;

generate, using said computer, a first classifier adapted to provide column correlation values reflecting the probability that a given column is said target column;

generate, using said computer, a second classifier adapted to provide row correlation values reflecting the probability that a given row is said target row;

apply, using said computer, said first classifier to the columns in the target table to determine a column correlation value for each column;

apply, using said computer, said second classifier to the rows in the target table to determine a row correlation value for each row;

suggest, using said computer, as the target cell, a cell having elevated column and row correlation values relative to other cells in the target table;

receive, by said computer, a set of training data; and

fine-tune, by said computer using said training data, said first classifier and said second classifier.

16. The computer program product of claim 15 , wherein said method further comprises:

the receiving, using said computer, of the set of training data including a training question, a training table, and identification of a ground truth target answer cell location within said training table, said target cell location corresponding to a target column and a target row;

the fine-tune, using said computer, of said first classifier to provide column correlation values using said training data; and

the fine-tune, using said computer, of said second classifier to provide row correlation values using said training data.

17. The computer program product of claim 16 , wherein said fine tuning is conducted, using said computer, at least on part with weak supervision 3{3,10,17} of {2,9,16}.

18. The computer program product of claim 15 , wherein said target cell has the highest row relevance value within a group of cells located in a column having a highest column correlation value.

19. The computer program product of claim 15 , wherein said classifier is a pre-trained transformer-model adapted to execute a next sentence prediction task.

20. The computer program product of claim 15 , wherein said classifier is a linear regression model adapted to classify feature vectors.