dc.description.abstract | There is a growing number of notifications generated from a wide range of sources. However, to our knowledge, there is no well-known generalizable standard for detecting the most urgent notifications. Establishing reusable standards is crucial for applications in which the recommendation (notification) is critical due to the level of urgency and sensitivity (e.g. medical domain). To tackle this problem, this thesis aims to establish Information Retrieval (IR) standards for notification (recommendation) task by taking semantic dimensions (terms, opinions, concepts and user interaction) into consideration. The technical research contributions of this thesis include but not limited to the development of a semantic IR framework based on Dirichlet Compound Model (DCM); namely FDCM, extending FDCM to the recommendation scenario (RFDCM) and proposing novel opinion-aware ranking models. Transparency, explainability and generalizability are some benefits that the use of a mathematically well-defined solution such as DCM offers. The FDCM framework is based on a robust aggregation parameter which effectively combines the semantic retrieval scores using Query Performance Predictors (QPPs). Our experimental results confirm the effectiveness of such approach in recommendation systems and semantic retrieval. One of the main findings of this thesis is that the concept-based extension (term-only + concept-only) of FDCM consistently outperformed both terms-only and concept-only baselines concerning biomedical data. Moreover, we show that semantic IR is beneficial for collaborative filtering and therefore it could help data scientists to develop hybrid and consolidated IR systems comprising content-based and collaborative filtering aspects of recommendation. | en_US |