Understanding LLMSynonymRetriever's node scoring and max_depth parameter
Understanding LLMSynonymRetriever's node scoring and max_depth parameter
At a glance
The community member is trying to use the LLMSynonymRetriever to check for existing nodes in their graph and reuse them instead of creating new ones. They are having trouble understanding how the retriever scores the nodes and the max_depth parameter, especially since the new node they want to insert is not connected to the graph.
In the comments, another community member explains that there is no scoring, it's just an exact match, and the max_depth parameter refers to how deep the retriever should look for related triplets (i.e., how many neighbors to travel to). The idea behind the retriever is to take a natural language input and use it for retrieval, whether that's through vector similarity, generating synonyms, or other methods.
There is no explicitly marked answer in the provided information.
Hi everyone , I am trying to utilise LLMSynonymRetriever in order to check for existing nodes in my graph and if they do not create a new node but rather utilise an existing one. I am having a hard understanding how it scores the nodes and also the parameter for max_depth , my question is the new node that is about to be inserted is not connected to the graph, so how does it determine how deep to walk into the graph since the node is not connected ?
hi Logan , thanks very much for your answers , much appreciated!One question on your response. So the idea behind the retriever is to compare an already existing node with other nodes in the graph ?