Fork me on GitHub

do you know if knowledge graphs are useful or its too complicated and AI to be used yet?


Yes, they are useful, but in different ways. Knowledge Graphs encode what you know about the data. AI attempts to determine what data means. They are used for different tasks. The trick is in getting them to work together, since it would be great for an AI to inform it’s decisions based on a collection of knowledge from a graph


i see so many data science jobs using statistics/probability etc but i like semantics


also in data engineering jobs, SQL is heavily used, like SQL data warehousing


also i like Mongodb and json-ld exists, i was wondering if i can use mongodb as triplestore also


This sort of architecture is possible; but as quoll says the database type determines what will be efficient/easy. For example you could put JSON documents in mongo, and provide a JSONLD interpretation of them also — but you’d want to make sure that when in mongodb you only ever stored/retrieved them in the ways that mongo is sympathetic too. If you wanted to then also query more flexibly as triples, you could ingest those same documents as triples into a triple store and query there. In this sense JSON-LD can act as an integration technology; i.e. giving universal meaning to localised data; but the difficulties will then be in determining which database is the source of truth etc — and how you can keep them in sync. You may find that the need to store/query in certain ways aligns with different audiences… e.g. there is a classical divide between transaction processing and analytical processing. So you can partition the ETL/synchronisation strategies accordingly; and things like JSON-LD can give you universal ways to reference data across both systems etc. I’d also state that I believe these integration benefits can be huge; but that they can be quite hard to achieve — not least because integration project teams often don’t align with the groups in charge of the individual systems. i.e. conways law is a barrier to what you can achieve; and the right approach is then a political process to agree to integrate in this manner. If it’s greenfield and you’re designing it up front; then that is obviously a lot easier to mandate these things — but you’ll be paying the costs of the integration and modelling — before you need it. So the challenge then becomes that people don’t see any benefits of integration until the different systems are independently useful; and it can be perceived as a pointless cost on getting things done.


Or rather it can be perceived as important; but not urgent — and therefore deprioritised — which leaves you with a political problem later 🙂


thank you for infomation rick, i was thinking that i would be nice to have data in JSON in localized way, and with JSON-LD to make them part of general knowledge. i will see how it goes in the future : )

👍 1

You can also of course apply context later; behind the scenes even… i.e. convert arbitrary JSON into JSON-LD (and triples) later; and people upstream of that need never know.