Date of Degree
domain-specific search, ontology engineering, web document re-ranking, ontology hierarchy extension, supervised learning, search framework
The goal of this research project is the realization of an artificial intelligence-driven lightweight domain knowledge search framework that returns a domain knowledge structure upon request with highly relevant web resources via a set of domain-centric re-ranking algorithms and adaptive ontology learning models. The re-ranking algorithm, a necessary mechanism to counter-play the heterogeneity and unstructured nature of web data, uses augmented queries and a hierarchical taxonomic structure to get further insight into the initial search results obtained from credited generic search engines. A semantic weight scale is applied to each node in the ontology graph and in turn generates a matrix of aggregated link relation scores that is used to compute the likely semantic correspondence between nodes and documents. Bootstrapped with a light-weight seed domain ontology, the theoretical platform focuses on the core back-end building blocks, employing two supervised automated learning models as well as semi-automated verification processes to progressively enhance, prune, and inspect the domain ontology to formulate a growing, up-to-date, and veritable system.\\ The framework provides an in-depth knowledge search platform and enhances user knowledge acquisition experience. With minimum footprint, the system stores only necessary metadata of possible domain knowledge searches, in order to provide fast fetching and caching. In addition, the re-ranking and ontology learning processes can be operated offline or in a preprocessing stage, the system therefore carries no significant overhead at runtime.
Zhao, Grace G., "Domain-Specific Knowledge Exploration with Ontology Hierarchical Re-Ranking
and Adaptive Learning and Extension" (2018). CUNY Academic Works.