Hi,
I’m currently using hybrid search which seems to work well but for my use case, I need to give documents which a more recent date a bit of a push.
I could implement this after retrieving the results but this feels hacky.
Is there maybe a way to use reranking for this? I have a date field but I’m not sure how I’d formulate a query that just boosts recent docs, not filter.
Thanks for weaviate, so far it’s awesome!
Hi, I’m not aware of a direct method that would do this unfortunately.
A janky method would be to concatenate a date dimension to your vector but that will probably get challenging to tune not to mention maintain. You might be also able to group results by age, if you have a coarse age column (like by month, or year).
What do you think @CShorten - any ideas?
Thanks for the tag @jphwang, love this topic! I think adding a Boost filter would be very similar in logic to our rerankers – definitely a nice stepping stone for symbolic rerankers (e.g. XGBoost models that take like product color, brand, so on…).
I will circle back here later this evening, hopefully with a GitHub issue to point you both towards haha – then we could rally the interest from there.
I asked for a similar “decay scoring”: [FR] Sort on near* for E-Commerce · Issue #3276 · weaviate/weaviate · GitHub
