-
Notifications
You must be signed in to change notification settings - Fork 2
ARAGORN
ARAGORN is a fully-federated knowledge graph querying and reasoning tool used to generate answers to medical research questions. ARAGORN communicates via the JSON-based Translator Reasoner API (TRAPI). Using this, ARAGORN can analyze a query graph, construct a knowledge graph related to that query graph, and then generate and score results from the constructed knowledge graph.
ARAGORN is comprised of multiple sub-services that each handle their own role in the query process:
- Strider
- Answer Coalesce
- ARAGORN Ranker
ARAGORN passes the query along to these separate sub-services, while also relying on a selection of external services as well, such as Node Normalization, as well as Knowledge Providers through Strider.
ARAGORN does accomplish a selection of tasks on its own. ARAGORN is responsible for query augmentation for creative queries, using a rule-based mechanism to create subqueries that correspond to the initial query. ARAGORN then compiles all the results for the initial query, using the results from the subqueries, as well as the necessary auxiliary graphs needed.
Strider is the ARAGORN subservice responsible for querying knoweldge graphs. ARAGORN sends Strider a query in the form of a TRAPI message, then Strider analyzes the query, constructs a query plan consisting of knowledge providers for each edge of the query graph, sends one hop queries to the KPs in the query plan, creates a knowledge graph pertaining to the query graph, and then generates results by binding knowledge graph nodes and edges to the query graph.
Strider accesses all KPs that are registered on SmartAPI.
Documentation for Strider can be found here.
Answer Coalesce takes a completed query and analyzes the results given. It coalesces results based on property, graph, and/or ontology.
Documentation for Answer Coalesce can be found here.
Ranker is the ARAGORN module responsible for ranking and scoring results. Ranker also accepts a completed query. Ranker constructs the Omnicorp overlay, which is the dense graph of literature co-occurrence based on PubMed publications between all nodes in the knowledge graph. Ranker then uses this pairwise literature co-occurrence, as well as properties listed on the edges themselves, such as publication counts and p-values, to score each result. These scores are bound to be between 0 and 1, where a higher score is better.
Documentation for ARAGORN Ranker can be found here.
ARAGORN only contacts knowledge providers through Strider. The full list of KPs accessed can be found here