Skip to content

Latest commit

 

History

History
101 lines (66 loc) · 3.71 KB

customizing_this_solution.md

File metadata and controls

101 lines (66 loc) · 3.71 KB

CosmosAIGraph : Customizing this Solution

This page strives to guide the reader in implementing their own application, based on CosmosAIGraph, with their own data.

This list of steps was useful in the development of CosmosAIGraph itself, and customers are recommended to take similar steps.

Step 1: Define your Requirements

It is essential to start here. Define your primary requirements. Specifically, define the list of graph queries in your application. Put these into a list to identify each one by a shortname (i.e. - q1, 2, etc.).

Use of the CQRS pattern is very useful here.

Step 2: Survey and Scan your Data

Ask the question: "Does the data that I have available enable the answering of these graph/knowledge-graph queries?".

If so, proceed. If not, revisit the requirements and/or data sources.

What do we mean by scanning?

By "scanning" we mean writing a program to read your input data to identify its structure, attributes and their data types, and the relationships between the entities.

There are several purposes of scanning:

  • To ensure that it is readable and parsable

  • To ensure that it contains the relationships, or connections, that you expect

    • The entities are linkable by specific attribute values

  • To capture the metadata of your data:

    • Its structure
    • Attribute names and their datatypes
    • Names of the classes and the relationships between them
    • Accurate metadata like this can greatly accelerate your development process.

Step 3: Understand RDF graph technology, ontologies, and SPARQL

RDF technology uses the beautifully simple concept of triples. Each triple consists of a subject, predicate, and object. A RDF database is simply an Ontology (i.e. - schema) with an array of many triples.

Though it's an old book, this O'Reilly book was very useful in learning RDF, OWL, and SPARQL at the beginning of the development of CosmosAIGraph: https://www.oreilly.com/library/view/learning-sparql/9781449311285/

Other learning sources include the following from the W3C:

Step 4: Create your initial Web Ontology

Based on your requirements and data, create your own *.owl file. Define the classes and relationships pertinent to your system.

See file impl/app/ontologies/libraries.owl as a reference.

The owl file development process should be iterative.

Step 5: Create a Data Wrangling process to populate a Jena graph

Given your datasets, and your evolving OWL file, populate an in-memory graph, then serialize it to a *.ttl file.

Write a console application to load your graph from the OWL and nt file, and craft SPARQL queries that seek to answer the queries you identified in Step 1.

You may need to implement programming logic to execute multiple SPARQL queries to satisfy an individual use-case requirement.

Expect steps 4 and 5 to be iterative. Modify your OWL, refactor your data-wrangling process, and recreate your graph and nt file.

Step 6: Create a minimal "SPARQL QUERY" web interface

Only after you have a reasonably working graph and ontology should you try to port it to a web application.

It is recommended that you implement your own "SPARQL Query" UI, similar to the CosmosAIGraph implementation, so that you can further explore and iterate your OWL and graph.

Step 7: Create your web application UI

Once your OWL and graph design is stabilized, implement your desired UI.

This may include graph visualizations. This reference application uses the D3.js JavaScript library for visualizations, but of course you may use any libraries you prefer.