The current SentenceSplitter has some issues with excludedEmbedMetadataKeys. The TextNodes created after the split containing the whole content including all metadata entries as text.
I'm not fully sure how this is supposed to work, but my naive assumption would be that MetadataAwarTextSplitter should use node.getContent(MetadataMode.NONE) instead of node.getContent(MetadataMode.ALL)
The current SentenceSplitter has some issues with
excludedEmbedMetadataKeys
. TheTextNode
s created after the split containing the whole content including all metadata entries as text.I'm not fully sure how this is supposed to work, but my naive assumption would be that
MetadataAwarTextSplitter
should usenode.getContent(MetadataMode.NONE)
instead ofnode.getContent(MetadataMode.ALL)