Generating explanatory paths for predicted column annotations
Systems, methods, and non-transitory computer-readable media are disclosed for generating generate explanatory paths for column annotations determined using a knowledge graph and a deep representation learning model. For instance, the disclosed systems can utilize a knowledge graph to generate an explanatory path for a column label determination from a deep representation learning model. For example, the disclosed systems can identify a column and determine a label for the column using a knowledge graph (e.g., a representation of a knowledge graph) that includes encodings of columns, column features, relational edges, and candidate labels. Then, the disclosed systems can determine a set of candidate paths between the column and the determined label for the column within the knowledge graph. Moreover, the disclosed systems can generate an explanatory path by ranking and selecting paths from the set of candidate paths using a greedy ranking and/or diversified ranking approach.
1. A computer-implemented method comprising:
identifying a column within a digital dataset;
determining a label for the column utilizing a knowledge graph by:
encoding the column, one or more column features, one or more relational edges, and one or more candidate labels within the knowledge graph; and
determining the label for the column utilizing annotation distance scores between the column and the one or more candidate labels within the knowledge graph;
determining a set of paths from the column to the determined label, wherein the paths from the set of paths comprise relational edge sequences and corresponding cumulative distance scores between the column, one or more neighbor nodes, and the determined label within the knowledge graph;
generating an explanatory path from the column to the determined label by selecting a path from the set of paths based on the cumulative distance scores corresponding to the set of paths; and
providing, for display via a user interface of a client device, the determined label and the explanatory path.
2. The computer-implemented method of claim 1 , further comprising generating the explanatory path from the column to the determined label by selecting a path corresponding to a lowest cumulative distance score from the cumulative distance scores corresponding to the set of paths.
3. The computer-implemented method of claim 1 , further comprising generating the explanatory path from the column to the determined label by selecting one or more paths from the set of paths based on the cumulative distance scores and relational edge sequence diversities corresponding to the set of paths.
4. The computer-implemented method of claim 1 , wherein the explanatory path comprises a representation of a relational edge sequence, corresponding to a selected path from the set of paths, between the column, one or more neighbor nodes from the selected path, and the determined label within the knowledge graph.
5. A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause a computer device to:
identify a column within a digital dataset;
determine a label for the column utilizing a knowledge graph by:
encoding the column, one or more column features, one or more relational edges, and one or more candidate labels within the knowledge graph; and
determining the label for the column utilizing annotation distance scores between the column and the one or more candidate labels within the knowledge graph;
determine a set of paths from the column to the determined label, wherein paths from the set of paths comprise relational edge sequences and corresponding cumulative distance scores between the column, one or more neighbor nodes, and the determined label within the knowledge graph;
generate an explanatory path from the column to the determined label by selecting a path from the set of paths based on the cumulative distance scores corresponding to the set of paths; and
provide, for display via a user interface of a client device, the determined label and the explanatory path.
6. The non-transitory computer-readable medium of claim 5 , further storing instructions that, when executed by the at least one processor, cause the computer device to generate the knowledge graph by encoding the one or more column features, the one or more relational edges, and the one or more candidate labels in a latent space.
7. The non-transitory computer-readable medium of claim 5 , further storing instructions that, when executed by the at least one processor, cause the computer device to determine the paths comprising the relational edge sequences and the corresponding cumulative distance scores between the column, the one or more neighbor nodes, and the determined label by:
identifying a first relational edge and a first distance value between the column and a neighbor node within the knowledge graph;
identifying a second relational edge and a second distance value between the neighbor node and the determined label within the knowledge graph;
determining a relational edge sequence utilizing the first relational edge and the second relational edge; and
determining a cumulative distance score utilizing the first distance value and the second distance value.
8. The non-transitory computer-readable medium of claim 5 , further storing instructions that, when executed by the at least one processor, cause the computer device to generate the explanatory path from the column to the determined label by selecting a path corresponding to a lowest cumulative distance score from the cumulative distance scores corresponding to the set of paths.
9. The non-transitory computer-readable medium of claim 5 , further storing instructions that, when executed by the at least one processor, cause the computer device to generate the explanatory path from the column to the determined label by selecting one or more paths from the set of paths based on the cumulative distance scores and relational edge sequence diversities corresponding to the set of paths.
10. The non-transitory computer-readable medium of claim 9 , further storing instructions that, when executed by the at least one processor, cause the computer device to select the one or more paths from the set of paths based on the cumulative distance scores and the relational edge sequence diversities corresponding to the set of paths by:
determining a first subset of paths corresponding to a first relational edge sequence pattern and a second subset of paths corresponding to a second relational edge sequence pattern;
selecting a first path from the first subset of paths corresponding to the first relational edge sequence pattern; and
selecting a second path from the second subset of paths corresponding to the second relational edge sequence pattern.
11. The non-transitory computer-readable medium of claim 10 , further storing instructions that, when executed by the at least one processor, cause the computer device to select the one or more paths from the set of paths based on the cumulative distance scores and the relational edge sequence diversities corresponding to the set of paths by:
ranking the first subset of paths and the second subset of paths based on the cumulative distance scores;
selecting the first path from the first subset of paths corresponding to the first relational edge sequence pattern based on the ranking of the first subset of paths; and
selecting the second path from the second subset of paths corresponding to the second relational edge sequence pattern based on the ranking of the second subset of paths.
12. The non-transitory computer-readable medium of claim 5 , further storing instructions that, when executed by the at least one processor, cause the computer device to generate the explanatory path from the column to the determined label by selecting a threshold number of paths from the set of paths based on the cumulative distance scores corresponding to the set of paths.
13. The non-transitory computer-readable medium of claim 5 , wherein the explanatory path comprises a representation of a relational edge sequence, corresponding to the selected path, between the column, one or more neighbor nodes from the selected path, and the determined label within the knowledge graph.
14. The non-transitory computer-readable medium of claim 5 , further storing instructions that, when executed by the at least one processor, cause the computer device to provide, for display via the user interface of the client device, the explanatory path indicating multiple paths to the determined label.
15. A system comprising:
one or more memory devices comprising a knowledge graph comprising a column, one or more column features, one or more relational edges, and one or more candidate labels; and
one or more server devices configured to cause the system to:
determine a label for the column utilizing the knowledge graph by:
encoding the column, one or more column features, one or more relational edges, and one or more candidate labels within the knowledge graph; and
determining the label for the column utilizing annotation distance scores between the column and the one or more candidate labels within the knowledge graph;
determine a set of paths from the column to the determined label, wherein paths from the set of paths comprise relational edge sequences and corresponding cumulative distance scores between the column, one or more neighbor nodes, and the determined label within the knowledge graph;
determine relational edge sequence diversities corresponding to the set of paths based on one or more relational edge sequence patterns corresponding to the paths from the set of paths;
generate an explanatory path from the column to the determined label by selecting a plurality of paths from the set of paths based on the cumulative distance scores and the relational edge sequence diversities corresponding to the set of paths; and
provide, for display via a user interface of a client device, the determined label and the explanatory path.
16. The system of claim 15 , wherein the one or more server devices are configured to cause the system to determine the label for the column utilizing the knowledge graph by utilizing annotation distance scores between the column and the one or more candidate labels within the knowledge graph, wherein the knowledge graph comprises an encoding of the column, the one or more column features, the one or more relational edges, and the one or more candidate labels in a latent space.
17. The system of claim 15 , wherein the one or more server devices are configured to cause the system to determine the paths comprising the relational edge sequences and the corresponding cumulative distance scores between the column, the one or more neighbor nodes, and the determined label by:
identifying a first relational edge and a first distance value between the column and a neighbor node within the knowledge graph;
identifying a second relational edge and a second distance value between the neighbor node and the determined label within the knowledge graph;
determining a relational edge sequence utilizing the first relational edge and the second relational edge; and
determining a cumulative distance score utilizing the first distance value and the second distance value.
18. The system of claim 15 , wherein the one or more server devices are configured to cause the system to select the plurality of paths from the set of paths based on the cumulative distance scores and the relational edge sequence diversities corresponding to the set of paths by:
determining a first subset of paths corresponding to a first relational edge sequence pattern and a second subset of paths corresponding to a second relational edge sequence pattern;
selecting a first path from the first subset of paths corresponding to the first relational edge sequence pattern; and
selecting a second path from the second subset of paths corresponding to the second relational edge sequence pattern.
19. The system of claim 18 , wherein the one or more server devices are configured to cause the system to select the plurality of paths from the set of paths based on the cumulative distance scores and the relational edge sequence diversities corresponding to the set of paths by:
ranking the first subset of paths and the second subset of paths based on the cumulative distance scores;
selecting the first path from the first subset of paths corresponding to the first relational edge sequence pattern based on the ranking of the first subset of paths; and
selecting the second path from the second subset of paths corresponding to the second relational edge sequence pattern based on the ranking of the second subset of paths.
20. The system of claim 15 , wherein the explanatory path comprises a representation of a relational edge sequence, corresponding to the selected path, between the column, one or more neighbor nodes from the selected path, and the determined label within the knowledge graph.