Connecting an Internal Chat Bot to Unstructured Data for Question Answering in Investment Research
A global investment bank’s research team needed a way to search through thousands of internal and external reports to provide client guidance.
These documents have ambiguous phrases, like “driving performance,” with specific meanings in the financial domain.
The Lymba Solution
Use specific ontologies to process research reports to create a natural language question-answering interface.
Process
We converted documents via semantics to RDF, stored them to an RDF Triple Store, and used NL2Query to retrieve data.
Outcome
This bank’s Investment Research team was able to ask questions of their data such as: “How did HPQ gross margins compare to the street?” To get the results they needed: “HP’s 1Q gross margin of 17.6% was down 500bps Y/Y and 800bps Q/Q, missing our forecast by -800bps and consensus by -700bps, falling to the lowest quarterly level since its separation from Hewlett Packard Enterprise in 2015.”