Dynamic question generation for information-gathering
Computer-based generation of information-gathering questions in response to a user query can include parsing the user query using natural language processing and extracting from the user query one or more phrases corresponding to a predetermined category. A knowledge database can be accessed, and entities semantically related to each phrase can be extracted therefrom. A query sub-graph representing a relationship between each of the one or more phrases and the entities extracted from the knowledge database can be generated. An expanded user query can be generated by traversing the query sub-graph. Passages semantically related to the expanded user query can be retrieved from one or more passages databases and ranked. A neural question generator can generate a set of information-gathering questions based on the expanded user query and a group of select passages selected from the plurality of passages in accordance with each selected passage's ranking.
1. A computer-implemented method, the computer-implemented method comprising:
extracting one or more phrases from a user query, wherein the one or more phrases correspond to a predetermined category and are identified by parsing the user query using natural language processing;
accessing a predetermined knowledge database and extracting entities therefrom, wherein the entities extracted are semantically related to the one or more phrases;
generating a query sub-graph representing a relationship between each of the one or more phrases and the entities extracted from the predetermined knowledge database;
traversing the query sub-graph, and based on the traversing, generating an expanded user query;
retrieving from a passages database a plurality of passages that are semantically related to the expanded user query and ranking each of the plurality of passages based on semantic relatedness of each retrieved passage to the expanded user query; and
outputting a set of information-gathering questions generated with a neural question generator based on the expanded user query and a group of select passages, wherein each select passage of the group is selected from the plurality of passages based on a ranking of the passage.
2. The computer-implemented method claim 1 , further comprising:
re-ranking each of the plurality of passages to determine which of the plurality of passages to include in the group of selected passages.
3. The computer-implemented method of claim 2 , wherein the re-ranking is based on a proxy for measuring importance of each passage of the plurality of passages to the generating the set of information-gathering questions.
4. The computer-implemented method of claim 1 , wherein the neural question generator is configured as part of a generative adversarial network that also includes a neural question evaluator.
5. The computer-implemented method of claim 4 , wherein the neural question evaluator is configured to fine tune the neural question generator by identifying which of multiple predetermined criteria each information-gathering question satisfies and to generate rewards and losses accordingly using a reward function and a loss function.
6. The method of claim 1 , further comprising:
ordering the set of information-gathering questions using a conditional loss function.
7. The computer-implemented method of claim 1 , further comprising:
receiving a user response to at least one information-gathering question from the set of information-gathering questions;
identifying a conversational cue from the user response; and
updating the query sub-graph based on the conversational cue for an adaptive conversation.
8. The computer-implemented method of claim 1 , further comprising:
identifying at least one unanswerable question among the set of information-gathering questions to a user; and
conveying a prompt to the user via a user device requesting the user to provide a contextual cue relating to the at least one unanswerable question.
9. A system, comprising:
a processor configured to initiate operations including:
extracting one or more phrases from a user query, wherein the one or more phrases correspond to a predetermined category and are identified by parsing the user query using natural language processing;
accessing a predetermined knowledge database and extracting entities therefrom, wherein the entities extracted are semantically related to the one or more phrases;
generating a query sub-graph representing a relationship between each of the one or more phrases and the entities extracted from the predetermined knowledge database;
traversing the query sub-graph, and based on the traversing, generating an expanded user query;
retrieving from a passages database a plurality of passages that are semantically related to the expanded user query and ranking each of the plurality of passages; and
outputting a set of information-gathering questions generated with a neural question generator based on the expanded user query and a group of select passages, wherein each select passage of the group is selected from the plurality of passages based on a ranking of the passage, and wherein the neural question generator is configured as part of a generative adversarial network that also includes a neural question evaluator.
10. The system of claim 9 , wherein the processor is configured to initiate operations comprising:
ranking each of the plurality of passages retrieved from the passages database based on each retrieved passage's semantic relatedness to the expanded user query.
11. The system of claim 10 , wherein the processor is configured to initiate operations comprising:
re-ranking each of the plurality of passages to determine which of the plurality of passages to include in the group of selected passages.
12. The system of claim 11 , wherein the re-ranking is based on a proxy for measuring importance of each passage of the plurality of passages to the generating the set of information-gathering questions.
13. The system of claim 9 , wherein the neural question evaluator configured to fine tune the neural question generator by identifying which of multiple predetermined criteria each information-gathering question satisfies and to generate rewards and losses accordingly using a reward function and a loss function.
14. The system of claim 9 , wherein the processor is configured to initiate operations comprising:
ordering the set of information-gathering questions using a conditional loss function.
15. The system of claim 9 , wherein the processor is configured to initiate operations comprising:
receiving a user response to at least one information-gathering question from the set of information-gathering questions;
identifying a conversational cue from the user response; and
updating the query sub-graph based on the conversational cue for an adaptive conversation.
16. The system of claim 9 , wherein the processor is configured to initiate operations comprising:
identifying at least one unanswerable question among the set of information-gathering questions to a user; and
conveying a prompt to the user via a user device requesting the user to provide a contextual cue relating to the at least one unanswerable question.
17. A computer program product, comprising:
one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, wherein the program instructions are executable by computer hardware to initiate operations including:
extracting one or more phrases from a user query, wherein the one or more phrases correspond to a predetermined category and are identified by parsing the user query using natural language processing;
accessing a predetermined knowledge database and extracting entities therefrom, wherein the entities extracted are semantically related to the one or more phrases;
generating a query sub-graph representing a relationship between each of the one or more phrases and the entities extracted from the predetermined knowledge database;
traversing the query sub-graph, and based on the traversing, generating an expanded user query;
retrieving from a passages database a plurality of passages that are semantically related to the expanded user query and ranking each of the plurality of passages; and
outputting a set of information-gathering questions generated with a neural question generator based on the expanded user query and a group of select passages, wherein each select passage of the group is selected from the plurality of passages based on a ranking of the passage; and
ordering the set of information-gathering questions using a conditional loss function.
18. The computer program product of claim 17 , wherein the neural question generator is configured as part of a generative adversarial network that also includes a neural question evaluator configured to fine tune the neural question generator by identifying which of multiple predetermined criteria each information-gathering question satisfies and to generate rewards and losses accordingly using a reward function and a loss function.