Automated few-shot learning techniques for artificial intelligence-based query answering systems
Methods, systems, and computer program products for automated few-shot learning techniques for artificial intelligence-based query answering systems are provided herein. A computer-implemented method includes obtaining multiple sets of queries and answers associated with one or more tables; determining a level of complexity attributed to at least a portion of the queries from the multiple sets of queries and answers; determining, based at least in part on the determined level of complexity attributed to the at least a portion of the queries, one or more new queries for use in training at least one artificial intelligence-based query answering system; facilitating annotation of the one or more new queries; training the at least one artificial intelligence-based query answering system using at least a portion of the one or more annotated new queries; and performing at least one automated action using the at least one trained artificial intelligence-based query answering system.
1 . A computer-implemented method comprising:
obtaining multiple sets of queries and answers associated with one or more tables, wherein obtaining multiple sets of queries and answers associated with one or more tables comprises: (i) obtaining queries, derived from processing one or more query logs, pertaining to at least a first set of two or more differently structured tables, (ii) obtaining queries, generated by one or more users, pertaining to at least a second set of two or more differently structured tables, and (iii) obtaining queries, pertaining to at least a third set of two or more differently structured tables, by processing one or more table query-related data sources using at least one curriculum learning technique;
determining a level of complexity attributed to at least a portion of the queries from the multiple sets of queries and answers;
determining, based at least in part on the determined level of complexity attributed to the at least a portion of the queries and performance by at least one artificial intelligence-based query answering system on one or more existing portions of training data, one or more new queries for use in training the at least one artificial intelligence-based query answering system, wherein the at least one artificial intelligence-based query answering system comprises at least one neural network-based query answering system;
facilitating annotation of the one or more new queries;
training the at least one artificial intelligence-based query answering system using at least a portion of the one or more annotated new queries, wherein training the at least one artificial intelligence-based query answering system comprises implementing one or more few-shot learning techniques in connection with the at least a portion of the one or more annotated new queries; and
performing at least one automated action using the at least one trained artificial intelligence-based query answering system;
wherein the method is carried out by at least one computing device.
2 . The computer-implemented method of claim 1 , wherein using one or more artificial intelligence techniques comprises using at least one self-supervision query generation technique.
3 . The computer-implemented method of claim 1 , wherein facilitating annotation of the one or more new queries comprises obtaining annotations of the one or more new queries from at least a portion of the one or more users.
4 . The computer-implemented method of claim 1 , wherein performing at least one automated action comprises automatically answering one or more queries pertaining to at least a portion of the one or more tables using the at least one trained artificial intelligence-based query answering system.
5 . The computer-implemented method of claim 1 , wherein obtaining multiple sets of queries and answers associated with one or more tables comprises obtaining training data pertaining to each of multiple tables, wherein the training data for each of the multiple tables comprises a set of table-specific queries and a set of corresponding answers.
6 . The computer-implemented method of claim 1 , wherein facilitating annotation of the one or more new queries comprises automatically annotating at least a portion of the one or more new queries using at least one automated software program.
7 . The computer-implemented method of claim 1 , wherein software implementing the method is provided as a service in a cloud environment.
8 . A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computing device to cause the computing device to:
obtain multiple sets of queries and answers associated with one or more tables, wherein obtaining multiple sets of queries and answers associated with one or more tables comprises: (i) obtaining queries, derived from processing one or more query logs, pertaining to at least a first set of two or more differently structured tables, (ii) obtaining queries, generated by one or more users, pertaining to at least a second set of two or more differently structured tables, and (iii) obtaining queries, pertaining to at least a third set of two or more differently structured tables, by processing one or more table query-related data sources using at least one curriculum learning technique;
determine a level of complexity attributed to at least a portion of the queries from the multiple sets of queries and answers;
determine, based at least in part on the determined level of complexity attributed to the at least a portion of the queries and performance by at least one artificial intelligence-based query answering system on one or more existing portions of training data, one or more new queries for use in training the at least one artificial intelligence-based query answering system, wherein the at least one artificial intelligence-based query answering system comprises at least one neural network-based query answering system;
facilitate annotation of the one or more new queries;
train the at least one artificial intelligence-based query answering system using at least a portion of the one or more annotated new queries, wherein training the at least one artificial intelligence-based query answering system comprises implementing one or more few-shot learning techniques in connection with the at least a portion of the one or more annotated new queries; and
perform at least one automated action using the at least one trained artificial intelligence-based query answering system.
9 . The computer program product of claim 8 , wherein using one or more artificial intelligence techniques comprises using at least one self-supervision query generation technique.
10 . The computer program product of claim 8 , wherein facilitating annotation of the one or more new queries comprises obtaining annotations of the one or more new queries from at least a portion of the one or more users.
11 . The computer program product of claim 8 , wherein performing at least one automated action comprises automatically answering one or more queries pertaining to at least a portion of the one or more tables using the at least one trained artificial intelligence-based query answering system.
12 . The computer program product of claim 8 , wherein obtaining multiple sets of queries and answers associated with one or more tables comprises obtaining training data pertaining to each of multiple tables, wherein the training data for each of the multiple tables comprises a set of table-specific queries and a set of corresponding answers.
13 . The computer program product of claim 8 , wherein facilitating annotation of the one or more new queries comprises automatically annotating at least a portion of the one or more new queries using at least one automated software program.
14 . A system comprising:
a memory configured to store program instructions; and
a processor operatively coupled to the memory to execute the program instructions to:
obtain multiple sets of queries and answers associated with one or more tables, wherein obtaining multiple sets of queries and answers associated with one or more tables comprises: (i) obtaining queries, derived from processing one or more query logs, pertaining to at least a first set of two or more differently structured tables, (ii) obtaining queries, generated by one or more users, pertaining to at least a second set of two or more differently structured tables, and (iii) obtaining queries, pertaining to at least a third set of two or more differently structured tables, by processing one or more table query-related data sources using at least one curriculum learning technique;
determine a level of complexity attributed to at least a portion of the queries from the multiple sets of queries and answers;
determine, based at least in part on the determined level of complexity attributed to the at least a portion of the queries and performance by at least one artificial intelligence-based query answering system on one or more existing portions of training data, one or more new queries for use in training the at least one artificial intelligence-based query answering system, wherein the at least one artificial intelligence-based query answering system comprises at least one neural network-based query answering system;
facilitate annotation of the one or more new queries;
train the at least one artificial intelligence-based query answering system using at least a portion of the one or more annotated new queries, wherein training the at least one artificial intelligence-based query answering system comprises implementing one or more few-shot learning techniques in connection with the at least a portion of the one or more annotated new queries; and
perform at least one automated action using the at least one trained artificial intelligence-based query answering system.
15 . The system of claim 14 , wherein using one or more artificial intelligence techniques comprises using at least one self-supervision query generation technique.
16 . The system of claim 14 , wherein facilitating annotation of the one or more new queries comprises obtaining annotations of the one or more new queries from at least a portion of the one or more users.
17 . The system of claim 14 , wherein performing at least one automated action comprises automatically answering one or more queries pertaining to at least a portion of the one or more tables using the at least one trained artificial intelligence-based query answering system.
18 . The system of claim 14 , wherein obtaining multiple sets of queries and answers associated with one or more tables comprises obtaining training data pertaining to each of multiple tables, wherein the training data for each of the multiple tables comprises a set of table-specific queries and a set of corresponding answers.
19 . The system of claim 14 , wherein facilitating annotation of the one or more new queries comprises automatically annotating at least a portion of the one or more new queries using at least one automated software program.