Closed-loop natural language query pre-processor and response synthesizer architecture
A closed-loop natural language query pre-processor and response synthesizer architecture accepts natural language queries and dynamically synthesizes query results. The query results may be in the form of data stories. The architecture identifies, selects, and composes candidate response elements into a coherent and meaningful query result. The architecture also implements an adaptable delivery mechanism that is responsive to connection bandwidth, query source preferences, query source characteristics, and other factors. Feedback from multiple sources adapts the architecture for handling subsequent queries.
1. A natural language query processing system comprising:
communication interface configured to connect to a query source and obtain a natural language input query from a requesting entity;
query pre-processing circuitry comprising:
a session controller configured to:
establish a query session for the query source in response to obtaining the natural language input query;
access a prior query history to identify prior query characteristics; and
submit the natural language input query to a question handler; and
a question handler configured to:
query an enterprise database for an enterprise linked to the requesting entity, to obtain requester metadata that identifies characteristics of the requesting entity;
identify query components comprising:
an explicit entity that is explicitly included in the natural language input query;
an implicit entity implied but not explicitly included in the natural language input query;
selected characteristics of the prior query characteristics; and
selected characteristics of the requesting entity in the requester metadata;
execute searches formulated with the query components against pre-defined data stores to determine candidate response elements; and
communicate the candidate response elements to response synthesis circuitry; and
response synthesis circuitry configured to:
perform assessments of the candidate response elements on a multi-dimensional analysis framework comprising a ‘relevance’ dimension and an ‘expected engagement’ dimension;
evaluate a function of the assessments on the ‘relevance’ dimension and the ‘expected engagement’ dimension to obtain an overall rank for each of the candidate response elements;
filter the candidate response elements by their overall rankings against a configurable relevance cut-off to obtain a filtered element subset;
determine a presentation sequence for the candidate response elements in the filtered element subset;
determine a query response package structure;
synthesize a query response to the natural language query input by:
ordering the candidate response elements in the filtered element subset according to the presentation sequence; and
generating content elements for the query response from the candidate response elements in the filtered element subset according to the query response package structure; and
deliver the query response to the query source via the session controller and the communication interface.
2. The system of claim 1 , where:
the requester metadata comprises any combination of one or more of:
a requesting entity role;
a requesting entity query response preference; and
a requesting entity enterprise activity;
the pre-defined data stores comprise:
a metrics data store comprising numerical data;
a meta data content data store comprising audio, visual, and textual data;
an enterprise data store comprising enterprise specific data content for the enterprise linked to the requesting entity; and
the query response package structure comprises a defined sequence of information data types following in sequence according to the presentation sequence.