[FEATURE] - Question contextualisation for MCQ Dataset #83
Description
Problem
When generating questions from a text snippet, the question is ok if all you have if this text snippet to answer. However in most realistic scenario we are building to either test domain-specific knowledge or retrieval capabilities. The questions may lack context.
Example question without context:
- At what mission time and display cue is the OPS 3 secondary actuator check initiated?
Same question with context:
- In the Space Shuttle Entry Flight Procedures for OPS 3 deorbit and entry interface, at what mission time and on which display cue is the secondary actuator check initiated?
Describe the solution you'd like
We can leverage another column from the source dataset context which provide some high-level explain to contextualise the generated question to the desired level.
Considerations
It can be done at generation time or post-generation.
Example post generation:
- Update the questions to include some brief context, like if the person asking the question would be explaining the context in which the question arise very concisely. For instance this document is about ... {context} = the Entry Flight Procedures of the Space Shuttle (Deorbit Burn, Entry interface (EI) and terminal area energy management (TAEM) interface).