Knowledge graphs
Table of Contents
Data in Context
Graphs are the bread and butter of data analysis. Understanding how to interpret these visualizations and build them is key to your success. Our solutions that provide interconnected data enriched with semantics, helps your organization in complex decision-making.
Our Methodology
Identification of Things & Concepts
- Concept types
- Data Field headers
- Concept Instances
URI Conceptualization
- URI Schemes
- Access Methods
- URI Management
Extraction
- Concept Extraction
- Concept Type prediction
- F-Score 98%
URI Assignment
- Concept Mapping to URI
- URI Management
Maintenance
- URI Change Management (Merging, Redirect ect)
- Backward Compatibility
- RDF Management
Content Delivery
- RDF based modelling
- SPARQL Interface
- API systems for Integration
- Authentication Management
Linked Data & Ontologies
FAIR Data and processes
Presenting a use case, by adopting FAIR principles, how greater insights and better content usage experience can be created for the customer leveraging machine and deep learning. Fair data enhances discoverability, cross-talk, content repurposing and introduces the possibility of new revenue models for the business.
FAIR principles also made it easier and to efficiently include human-in-the-loop feedback and optimization, both at metadata level and process level as a whole for the entire application.
- How were FAIR principles adopted for both metadata and processes?
- What role did FAIR principles play in enhancing AI/ML efficiency?
- Is there a cost-benefit angle to this adoption? If yes, what is the learning?
Defining our interpretation of FAIR principles
Findable
Not about standards but about having the right metadata in the right format, with due diligence to flexibility in terms of enrichments and updates
Accessible
Implementation plans that aim at harmonization of content, metadata and applications
Interoperable
Exploiting integration ready core components together and/or as individual solutions
Findable
- Findability is achieved by using dedicated URIs and endpoints for each concept in the ontology, which in turn is used as the base to tag, classify content or attribute metadata. A detailed provenance is always ensured and maintained
- All and any value addition to content via ontologies and or other metadata is both human and machine accessible-readable-usable
- Processes such as search and retrieval employ findability as a principle by having minimum restrictions on the user with respect to actions required to infer/discover knowledge. This is achieved by defining a clear, unambiguous data model with semantic concepts and associations as basic units
Accessible
Accessibility is achieved by carving out specific rules for archival of metadata, content and processes
For instance, if an alert service is being opted for on specific clinical studies involving X intervention vs Y intervention, clear instructions are set on when to archive the content, if archived, what is made accessible if user has required permissions and in what format? (Say redirects to original content or summaries only etc, etc.,)
Machine interpreted standard licenses also form an important part of the accessibility strategy. Due diligence to web content accessibility
Interoperable
Interoperability is achieved in one way by defining metadata and associations in such a way that they can be leveraged within the workflow for different applications/user base
Another implementation is making the actual metadata interoperable by sticking to same standards. For instance a concept maybe a drug-chemical or a medical procedure. But the properties related to each are defined with a clear scope that makes it not only easy to differentiate but also to allow end user applications to interact with them in a similar fashion despite the semantic differences
Final application is in the way different contents are levelled in order to use same metadata to make them discoverable. For instance, handling differences in content types structurally, format considerations like text, audio video etc.,
Reusable
Reusability is achieved by laying out heavily structured and documented steps for processes and applications
Another parallel to reusability employed wherein; the same metadata attribution module (ML based tagging) can be used to fingerprint and aid in searchability for end users and authors, while the same fingerprints are leveraged to allow automated flow of the content through applications be it a physician, researcher, marketing professional or a clinician. Further on, marketing decisions and strategic planning are also centered around the same fingerprints/metadata
Lastly, but rightly, scope for improvement/expansion is achieved by ensuring that a core is set up in a way that makes it practical to append to the existing content/metadata instead of a complete turnaround wherever applicable. Classical example is accomodating new drugs/drug candidates with every passing day or with every content upgrade. There is always a reusable component that is a major contributor in the solution provided
Measuring Acceptability and Success
- Ability to retrieve most relevant results with minimal input
- Ability to slice and dice content for knowledge discovery via “smart prompts”
- Flexibility to exploit the offering : no restrictions on knowing “precise keywords/terms/entities” to search for or explicitly make use of a search syntax
- Indiscriminate access to content of different types (text through videos)
- Allow feedback integration without any restrictions on the format/channels via which feedback is received (For instance most interacted vs least interacted recommendations, actual metadata attributed etc.,)
- Updates and switch overs to handle new content – TAT
- Concurrent user accommodations w.r.t searches
- Discoverability on web and IoT (Google search accessibility and metrics)
Measuring Usability and Success
- Domain Specific taxonomy developed and semantic units attributed allow us to provide added value to users.
- The taxonomy facilitates concept searching rather than matching individual words.
- Semantic metadata especially binds the entire solution together, contributing to;
- Discoverability and search/retrieval
- Binding content across the offering (Text, video, audio)
- Creating ancillary offerings : Alerts, Recommendations, Advertisements
- Natural Language like query based email alerting system which is cross-journal and includes other content types (such as standards and guidelines).
- All metadata and standards used thereof contribute towards visibility, reusability, reproducibility and scalability of the offering
