Replies: 9 comments
|
what is the best configurations for my usecase ? each brand send 2-3 emails per day, I want to have a good memory so i can ask questions like "what are the channels supported by brand X", "what is the final budget of brand X" etc.. |
|
Hey <@714543164462989323> thanks for sharing your work. For your use case - first it makes sense to keep second when asking about current or final state you may want to use third maintaining separate datasets per brand ex - fourth regarding the |
|
Overall your ingestion logic appears fundamentally sound. the inconsistencies you are seeing are likely due to one of the following: using Using |
|
what about ontology, is it recommended over the DatePoints ? |
|
Ontology and datapoints serve different purposes. an ontology helps improve how information is extracted from your email text during dataPoints are useful when you already have structured data, such as parsed budget or channel fields. they allow you to insert that information directly into the graph, which gives you more precise and deterministic results. for your use case starting with temporal |
|
<@1373135538147692604> when using temporal_cognify=True on the cognify(), it seems silently ignores ontology and custom prompt: If temporal_cognify = False, is it still ok using SearchType.TEMPORAL on the search ?maybe with highlighting the importance of the dates in the custom prompt |
|
<@1373135538147692604> when using temporal_cognify=True on the cognify(), it seems silently ignores ontology and custom prompt: If temporal_cognify = False, is it still ok using SearchType.TEMPORAL on the search ?maybe with highlighting the importance of the dates in the custom prompt |
|
yep that's correct with |
|
if you still want to influence the answers a practical approach is to use a strict |
Uh oh!
There was an error while loading. Please reload this page.
=Hi Team,
I've been testing Cognee for a use case where I get daily emails from brands updating me on their campaigns — budget changes, supported marketing channels, that kind of thing. I wanted a way to persist this memory over time, but Cognee wasn't giving me accurate answers when I queried it. Here's an example of what I'm ingesting daily — any recommendations?
I attached a file on how my json looks like, and another file for my script, I have created 3 memories, one for each brand and one for all(for cross-join questions),
I used
cognify(temporal_cognify=True, incremental_loading=True)This discussion was automatically pulled from Discord.
All reactions