llama_index SPARQL Notes 05

Published on 2023-08-31 by @danja

Grr...next admin session I must set up systemd

sudo /usr/local/nebula/scripts/nebula.service start all

for ChatGPT :

I'd like a function to remove duplicates from a Python json structure. For example, give the following :
[
{'s': 'production on Vol.3', 'p': 'was put on hold until', 'o': 'February 2021'
},
{'s': 'production on Vol.3', 'p': 'put on hold until', 'o': 'February 2021'
},
{'s': 'production on Vol.3', 'p': 'was put on hold until', 'o': 'February 2021'
}
]
the function should return :
[
{'s': 'production on Vol.3', 'p': 'was put on hold until', 'o': 'February 2021'
},
{'s': 'production on Vol.3', 'p': 'put on hold until', 'o': 'February 2021'
}
]

It got it right first time!

So, next to flip this JSON into RDF/SPARQL.

Not 100% sure, but I don't think I actually need the extract_entities function, anything useful will also appear in extract_entities and the types/roles are implied by the triplets.

sanitize

def escape_for_rdf(input_str) - thank you ChatGPT.

Given the following JSON, how would I retrieve the values of s, p, o? [ {'s': "Industrial Light & Magic's StageCraft", 'p': 'was developed for', 'o': 'Disney+ Star Wars series The Mandalorian' } ]

took a little backwards & forwards, but we got there.

I'm really not sure what level of granularity will eventually be needed, for big lumps of data something else from the SPARQL Protocol would probably be better (like a full-on HTTP PUT). Per-triple DELETE/INSERT might be desirable elsewhere. But here it shouldn't get too big, the INSERT approach should work, and would also work per-triple.

The NebulaGraph notion of a Space appears very similar to RDF/SPARQL Named Graphs, so I'll go with that (more information, and things can get messy doing everything in the default graph).

This is what I came up with the other day :

# Simple Entity-Relation

@base <http://purl.org/stuff/data> .
@prefix er: <http://purl.org/stuff/er> .

<#T123> a er:Triplet ;
er:id "#T123" ;
er:subject <#E123> ;
er:property <#R456> ;
er:object <#E567> .

<#E123> a er:Entity ;
er:value "one" .

<#R456> a er:Relationship ;
er:value "two" .

<#E567> a er:Entity ;
er:value "three" .

I opted for URL-named resources rather than blank nodes or other IRI because soon it may help with sanity-checking, further down the line the potential for HTTP retrieval is nice to have. There are a lot of annotation/meta triples that could be added, but I reckon this is about the minimum necessary to fulfil the Wey Gu's Notebook with SPARQL use case.

I guess I need a fragment ID generator. Suboptimal but so it's not too hard on the eyes for now, I'll got with alpha(capitals)-numeric. How many chars?

Please write a function to generate a random 4-character string using only numeric characters and capital letters.

Ok.

Need to keep identity of entities/relationships. I'm not sure what should happen if the same string is pulled out more than once as entity and/or relationship from different contexts. Should they have different IDs (URLs)? I don't fancy a deep dive into llama_index structures right now, I'll wait for someone to tell me. Whatever, giving identical strings the same URL shouldn't make a huge difference either way.

Strings appearing both in the role of Relationship & Entity might also be thinking about. But again, for now, no big deal.

Ran current version -

                        <#THKOE> a er:Triplet ;
                                er:subject <#EK8WH> ;
                                er:property <#RJSJV> ;
                                er:object <#ELD8T> .

                        <#EK8WH> a er:Entity ;
                                er:value "Vol.3" .

                        <#RJSJV> a er:Relationship ;
                                er:value "is the first MCU film to feature" .

                        <#ELD8T> a er:Entity ;
                                er:value "the word  fuck  uncensored" .

Progress!