Adapting from 1-to-1 to 1-to-many

[EddieJ03]

I'm interested in augmenting this repo for an EM task, and was wondering if there are any suggestions on the best way to adapt this codebase to fit a 1-to-many scenario instead of 1-to-1. In other words, given an entity record $r$ and its set of potential matches, find all matches from the set.

[tshu-w]

Thanks for your interest in our work.
I think the following changes are needed to adapt from 1-to-1 to 1-to-many:

1. For the matching strategy, maybe nothing to do.
2. For the comparing strategy, change from top-1 to top-k validation, where the number of k's is related to the number of "many".
3. For the selecting strategy, modify the prompt to allow LLM to select multiple records and adjust the extraction result code is the easy way.

However, which strategy is more effective and how to combine them may need further exploration in the 1-to-many case.