Machine Learning and Ontologies

Whilst considering the benefits of ontology based data sources to ingest into H2O and the machine learning world, I came across two interesting papers:

  • Ontology Matching: A Machine Learning Approach
  • Using Machine Learning to Support Continuous Ontology Development

I’m not sure if either have helped me with my thought process, but they are still interesting.  In many ways, I’m wondering if Ontologies would aid in model/algorithm usage, or I end up with data overload 🙂

Its also worth having a look at “How Goldman Sachs is Using Knowledge to Create an Information Edge“.  Although the slide deck isn’t full of detail, slide 5 does at least off a high level architecture, and provide some detail on ontology usage. Looks like the same presentation was delivered again last year – unfortunately this time there isn’t any slide deck available.

In the Goldman approach, authoritative raw data sources are stored in an enterprise data lake in Hadoop, stored in HDFS. To represent our knowledge domains, Goldman Sachs has developed an enterprise ontology representing its business concepts. The upper ontology is represented in OWL format and describes class structures and data transformation rules. These rules are then leveraged in a custom-developed Hadoop process to transform data into semantically consistent information in RDF format. The assertions and higher-order concepts from the lower ontology are leveraged to generate the rule set on the transformed data, creating “Big Graphs” consisting of, in some cases, billions of nodes and edges


~ by mdavey on March 23, 2016.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: