Claims:WE CLAIM:

1. A method for generating question variations to user input, said method comprising:
receiving, by a response generation system, a user input comprising at least one sentence from a user;
extracting, by a response generation system, plurality of keywords and associated plurality of features from the at least one sentence;
generating, by a response generation system, a plurality of question variations for the user input by using one or more subgraphs identified from a trained knowledge graph based on the plurality of keywords and the associated plurality of features, wherein the trained knowledge graphs is generated by performing the steps of:
determining one or more questions from a dataset provided during a training phase;
determining an association between one or more keywords extracted from the one or more questions and a context between the one or more keywords, wherein the association between the one or more keywords represent at least one of an entity and an action, and the context between the one or more keywords are represented by one or more features; and
generating a knowledge graph comprising the one or more features, wherein each of the one or more features is associated with a confidence score, wherein the knowledge graph is used for generating one or more question variations for the dataset.

2. The method as claimed in claim 1, wherein the plurality of features may include at least one of word-based features and context-based features, wherein the word-based features comprises morphological features, semantics and syntax and the context-based features comprises hierarchical relations.

3. The method as claimed in claim 1, wherein the one or more subgraphs comprise at least one of an action specific graph and an entity specific graph, wherein the action specific graph denotes a link between an action word and possible entities associated with the action word and wherein the entity specific graph denotes a link between the entity and possible actions associated with the entity.

4. The method as claimed in claim 1, wherein the one or more questions are determined using reinforcement learning.

5. The method as claimed in claim 1, wherein the trained knowledge graph is generated using bidirectional LSTM (Long Short-Term Memory) and wherein the association is determined using spreading activation.

6. A question variation generation system, for generating question variations to user input, said system comprising:
a processor; and
a memory, communicatively coupled with the processor, storing processor executable instructions, which, on execution causes the processor to:
receive, a user input comprising at least one sentence from a user;
extract, plurality of keywords and associated plurality of features from the at least one sentence;
generate, a plurality of question variations for the user input by using one or more subgraphs identified from a trained knowledge graph based on the plurality of keywords and the associated plurality of features, wherein the trained knowledge graphs is generated by performing the steps of:
determining one or more questions from a dataset provided during a training phase;
determining an association between one or more keywords extracted from the one or more questions and a context between the one or more keywords, wherein the association between the one or more keywords represent at least one of an entity and an action, and the context between the one or more keywords are represented by one or more features; and
generating a knowledge graph comprising the one or more features, wherein each of the one or more features is associated with a confidence score, wherein the knowledge graph is used for generating one or more question variations for the dataset.

7. The question variation generation system as claimed in claim 6, wherein the plurality of features may include at least one of word-based features and context-based features, wherein the word-based features comprises morphological features, semantics and syntax and the context-based features comprises hierarchical relations.

8. The question variation generation system as claimed in claim 6, wherein the one or more subgraphs comprise at least one of an action specific graph and an entity specific graph, wherein the action specific graph denotes a link between an action word and possible entities associated with the action word and wherein the entity specific graph denotes a link between the entity and possible actions associated with the entity.

9. The question variation generation system as claimed in claim 6, wherein the one or more questions are determined using reinforcement learning.

10. The question variation generation system as claimed in claim 6, wherein the trained knowledge graph is generated using bidirectional LSTM (Long Short-Term Memory) and wherein the association is determined using spreading activation.

Dated this 31st day of July, 2018

R Ramya Rao
Of K&S Partners
Agent for the Applicant
IN/PA-1607
, Description:TECHNICAL FIELD
The present disclosure relates to the field of artificial intelligence. More particularly, but not exclusively, the present disclosure relates to a method and system for generating question variations.