Facilitating Data-Centric Recommendation in Knowledge Graph

In order to help Earth scientists share knowledge and expedite scientific exploration, NASA is developing NASA Earth Science Enterprise (ESE) whose underlying basis is NASA Science Knowledge Graph (SKG) that automatically extracts knowledge from scientific publications. The ultimate goal is to provide a one-stop gateway that is able to mine, all together, data-centric knowledge including datasets, tools and models, algorithms, statistical technology, projects, topics, hypotheses and conclusions, as well as hidden information from literature. Paper reviewers could leverage SKG to assess a submission in a peer-review process to ensure FAIR compliance.

To automate the development and enrichment of the SKG, we have developed an information model to carry key information entities and relationships around scientific data. Traditional entity- and relationship-encapsulated features are extracted as first-class citizens in our SKG, so that implicit feature relationships can be turned into explicit structural relationships in the SKG. As a result, we could leverage structural relationships embedded in SKG to explore hidden information at runtime, in contrast to semantic analysis where application-specific rules have to be manually coded. Based upon the information model and knowledge network constructed, we have developed a tailored intent-oriented, context-aware technique to answer community-oriented queries and provide personalized recommendation. Furthermore, we have applied deep learning algorithms to detect structural similarity and explore paths at runtime, respectively, for addressing scalability and performance issues.