Leomrlin commented on issue #773:
URL: https://github.com/apache/geaflow/issues/773#issuecomment-4028338418

   Hi, @SYaoJun 
   We greatly appreciate your interest in contributing to these vector 
implementations!
   Regarding MagnitudeVector: This vector type is designed to capture the LLM's 
sensitivity to magnitude or quantitative information embedded within memories. 
For instance, when a user specifies a timeframe like "between 2003 and 2023," 
this vector could guide the memory system to retrieve content associated with 
similar or overlapping numerical ranges. Another example would be recalling 
memories tagged with "temperature above 30°C" when such a condition is 
mentioned, effectively querying an open interval.
   Regarding TraversalVector: This vector aims to capture the LLM's 
understanding of graph semantics to represent graph traversal intents. For 
example, in a knowledge graph containing entity types like "Person" and 
"School" connected by a "graduated from" relation, the LLM could formulate a 
query vector such as (Person_X, graduated from, School). This would instruct 
the memory system to perform a structured recall based on the graph's 
connectivity patterns.
   Currently, this memory system is still in the active research and 
development phase. We are exploring and refining these concepts, and we warmly 
welcome you to join our open-source community to collaborate on building it 
out. Please also note that this is a preliminary draft—these ideas may not yet 
be fully mature, and we are very open to more compelling suggestions or 
alternative approaches.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to