Based on my understanding, the new pipeline should look like this (this example is for INSERT action/intent):
1) NL received by the backend
2) NL stored to KG as statement
3) Extract triple as string (subject, predicate, object, preference)
4) Link any entities found
5) Extract additional anno
5) Update the KG entry with this additional information.
I suggest a slight simplification. We start by creating three dataclasses:
@dataclass
class TripleAnnotation:
subject: str
predicate: Optional[str] = None
object: Optional[str] = None
@dataclass
class PreferenceAnnotation:
topic: Uri
weight: float
@dataclass
class KnowledgeData:
statement: string
triple: Optional[TripleAnnotation] = None
preference: Optional[PreferenceAnnotation] = None
# ... metadata
This way we can streamline the process of inserting, and only insert at the end when the dataclasses are populated. This also simplifies moving information between modules as we only need to pass KnowledgeData instead of all arguments separately.
One thing I am unsure about is how to extend this to add linked entities. Maybe another dataclass UriTripleAnnotation or similar could be used.
Based on my understanding, the new pipeline should look like this (this example is for INSERT action/intent): 1) NL received by the backend 2) NL stored to KG as
statement3) Extract triple as string (subject, predicate, object, preference) 4) Link any entities found 5) Extract additional anno 5) Update the KG entry with this additional information.I suggest a slight simplification. We start by creating three dataclasses:
This way we can streamline the process of inserting, and only insert at the end when the dataclasses are populated. This also simplifies moving information between modules as we only need to pass
KnowledgeDatainstead of all arguments separately.One thing I am unsure about is how to extend this to add linked entities. Maybe another dataclass
UriTripleAnnotationor similar could be used.