Method and system for electronic processing of user queries maintaining factual consistency during processing
A method for electronic processing of user queries maintaining factual consistency during processing includes receiving a user query, executing a hybrid retrieval operation, generating a context information space that includes a set of paragraph output from the paragraph retrieval operation, a set of triplet output from the triplet retrieval operation, and a set of graphs output from the graph retrieval operation. Furthermore, generating a first prompt instruction to generate an intermediate rationale as explanation output and performing an automatic fact-checking operation to verify the explanation output and re-feeding the user query along with the generated explanation output to the generative language model to derive an answer output that is succinct in one or more language parameters as compared to the generated explanation output. The answer output includes a verified and evidence-supported answer and at least one fact reference from first set of factual information with reduced hallucinations and factual inconsistencies.
1. A method for electronic processing of user queries maintaining factual consistency during processing, the method comprising:
receiving, by a server, a user query from a client device;
executing, by the server, a hybrid retrieval operation comprising a paragraph retrieval operation, a triplet retrieval operation, and a graph retrieval operation, to retrieve a first set of factual information based on the user query from a plurality of diverse data sources;
generating, by the server, a context information space comprising a set of paragraph output from the paragraph retrieval operation, a set of triplet output from the triplet retrieval operation, and a set of graphs output from the graph retrieval operation;
generating, by the server, a first prompt instruction comprising the user query as well as the retrieved first set of factual information in the form of the set of paragraph output, the set of triplet output, and the set of graphs output, to feed to a generative language model to generate an intermediate rationale as explanation output,
wherein each statement in the explanation output is linked to at least one fact reference from the first set of factual information as evidence;
performing, by the server, an automatic fact-checking operation to verify the explanation output against the retrieved first set of factual information; and
re-feeding, by the server, the user query along with the generated explanation output to the generative language model to derive an answer output that is succinct in one or more language parameters as compared to the generated explanation output,
wherein the answer output comprises a verified and evidence-supported answer and at least one fact reference from the first set of factual information, and
wherein the triplet retrieval operation comprises concurrently performing semantic encoding and lexical encoding of the user query and searching and extracting a set of triplets from a text embedding database based on the semantic encoding and the lexical encoding.
2. The method as claimed in claim 1 , further comprising a pre-processing stage comprising:
acquiring a dataset of documents from the plurality of diverse data sources; and
creating a knowledge graph database, a text embedding database, a node entity embedding database, and a triplet embedding database from the acquired dataset of documents based on a plurality of document processing operations.
3. The method as claimed in claim 2 , wherein the pre-processing stage further comprises indexing, caching, and creating parallel processing pathways for retrieval of information from the created knowledge graph database, the text embedding database, the node entity embedding database, and the triplet embedding database.
4. The method according to claim 1 , wherein the paragraph retrieval operation and the triplet retrieval operation are text-based retrieving operation that involves finding relevant and factual information present in the plurality of diverse data sources, whereas the graph retrieval operation is based on a graph-based database created by extracting information pieces from plurality of diverse data sources.
5. The method according to claim 1 , wherein the paragraph retrieval operation comprises:
concurrently performing semantic encoding and lexical encoding of the user query; and
searching and extracting a set of paragraphs from a semantic chunk and a textual chunk of a paragraphs embedding database.
6. The method according to claim 5 , wherein the paragraph retrieval operation further comprises re-ranking the extracted set of paragraphs to obtain a re-ordered set of paragraphs that corresponds to the set of paragraph output from the paragraph retrieval operation.
7. The method according to claim 1 , wherein the triplet retrieval operation further comprises re-ranking the extracted set of triplets to obtain a re-ordered set of triplets that corresponds to the set of triplet output from the triplet retrieval operation.
8. The method as claimed in claim 1 , wherein the graph retrieval operation comprises:
extracting a plurality of query entities from the user query;
searching the plurality of query entities in a node entity embedding database to retrieve one or more topic node entities; and
retrieve a set of sub-graphs from a knowledge graph database based on the one or more topic node entities and the user query.
9. The method as claimed in claim 8 , wherein the graph retrieval operation further comprises re-ranking the retrieved set of sub-graphs to obtain a re-ordered set of sub-graphs that corresponds to the set of graphs output from the graph retrieval operation.
10. The method according to claim 1 , further comprising assigning a fact identifier to each statement in the explanation output, wherein the fact identifier points to a fact taken from one of: the set of paragraph output, the set of triplet output, and the set of graphs output, to generate each statement in the explanation output.
11. The method according to claim 10 , wherein the fact identifier is one of: a unique paragraph identifier pointing to one paragraph of the set of paragraph output, a unique triplet identifier pointing to the set of triplet output, and a unique graph identifier pointing to the set of graphs output.
12. The method according to claim 1 , further comprising generating, by the server, a second prompt instruction comprising the user query as well as the generated intermediate rationale as the explanation output, for the re-feeding of the generative language model to generate the answer output.
13. The method according to claim 1 , further comprising performing an automatic grading of the answer output, wherein the automatic grading involves evaluating the answer output against the generated explanation output.
14. The method according to claim 1 , further comprises restricting the answer output based on the generated explanation output.
15. A system for electronic processing of user queries maintaining factual consistency during processing, the system comprising:
a server configured to:
receive a user query from a client device;
execute a hybrid retrieval operation comprising a paragraph retrieval operation, a triplet retrieval operation, and a graph retrieval operation, to retrieve a first set of factual information based on the user query from a plurality of diverse data sources;
generate a context information space comprising a set of paragraph output from the paragraph retrieval operation, a set of triplet output from the triplet retrieval operation, and a set of graphs output from the graph retrieval operation;
generate a first prompt instruction comprising the user query as well as the retrieved first set of factual information in the form of the set of paragraph output, the set of triplet output, and the set of graphs output, to feed to a generative language model to generate an intermediate rationale as explanation output,
wherein each statement in the explanation output is linked to at least one fact reference from the first set of factual information as evidence;
perform an automatic fact-checking operation to verify the explanation output against the retrieved first set of factual information; and
re-feed the user query along with the generated explanation output to the generative language model to derive an answer output that is succinct in one or more language parameters as compared to the generated explanation output,
wherein the answer output comprises a verified and evidence-supported answer and at least one fact reference from the first set of factual information, and
wherein the triplet retrieval operation comprises concurrently performing semantic encoding and lexical encoding of the user query and searching and extracting a set of triplets from a text embedding database based on the semantic encoding and the lexical encoding.
16. The system according to claim 15 , wherein the server is further configured to assign a fact identifier to each statement in the explanation output, wherein the fact identifier points to a fact taken from one of: the set of paragraph output, the set of triplet output, and the set of graphs output, to generate each statement in the explanation output.
17. The system according to claim 16 , wherein the fact identifier is one of: a unique paragraph identifier pointing to one paragraph of the set of paragraph output, a unique triplet identifier pointing to the set of triplet output, and a unique graph identifier pointing to the set of graphs output.
18. The system according to claim 15 , wherein the server is further configured to generate a second prompt instruction comprising the user query as well as the generated explanation output, for the re-feeding of the generative language model to generate the answer output.
19. The system according to claim 15 , wherein the server is further configured to restrict the answer output based on the generated explanation output.