580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can semantic modeling improve the transparency, relevance, and applicability of clinical decision support recommendations?
This research area investigates the use of semantic techniques to enrich clinical decision support (CDS) systems by formalizing guideline recommendations, provenance, and evidence applicability. It matters for enhancing trust, reducing alert fatigue, and supporting personalized patient care by making recommendations more explainable and contextually relevant.
Key finding: Introduced ontologies and semantic web applications to model diseases based on guidelines, capturing provenance and evidence specificity to improve the interpretability and applicability of CDS recommendations; demonstrated... Read more
Key finding: Developed a semantically-enabled platform integrated into personal health records that uses ontological annotations (SNOMED-CT, LOINC, RXTerms) to personalize medical information recommendations based on patient profiles and... Read more
Key finding: Proposed a semantic recommendation system using concept taxonomies to enhance user modeling via domain-based inference and recommendation via semantic similarity; experiments on Netflix movie data showed that semantic... Read more
2. What semantic techniques can enhance content-based recommender systems to better model user preferences and deliver diverse, serendipitous recommendations?
This direction explores the use of semantic technologies, ontologies, and reasoning to shift from keyword- or syntactic-based matching to concept-based representations of items and user profiles. It aims to overcome problems like overspecialization, limited content analysis, and ambiguity inherent in textual features, thereby enabling more accurate, interpretable, and serendipitous recommendations across domains.
Key finding: Reviewed semantic approaches that transform keyword-based profiles into concept-based representations using ontologies and open knowledge sources like DBpedia; identified top-down (ontology-driven) and bottom-up... Read more
Key finding: Demonstrated the use of semantic reasoning mechanisms including ontological hierarchies and spreading activation to infer complex semantic associations among items and dynamically learn evolving user preferences, thereby... Read more
Key finding: Implemented a system using high-level semantic descriptors derived from low-level audio features via SVM classifiers, enabling content-based music recommendation and preference visualization; demonstrated that semantic... Read more
3. How can hybrid semantic and collaborative filtering approaches mitigate data sparsity and cold-start problems in recommendation systems?
This theme focuses on integrating semantic information from ontologies or domain knowledge with collaborative filtering (CF) to address CF’s limitations such as sparse user-item rating matrices and new user/item cold starts. Hybridization techniques aim to build richer user profiles that incorporate semantic preferences, yielding improved recommendation accuracy and coverage.
Key finding: Proposed a hybrid recommendation system combining content-based filtering and collaborative filtering enhanced by semantic relationships and pattern mining to identify correlations between items and users; successfully... Read more
Key finding: Developed a novel hybridization technique (User Semantic Collaborative Filtering) that constructs user semantic profiles from structured item attributes and uses these to compute user similarity in a collaborative filtering... Read more
HCoF: Hybrid Collaborative Filtering Using Social and Semantic Suggestions for Friend Recommendation
Key finding: Introduced a hybrid collaborative filtering approach incorporating social behavior measures (friendship, trust, commitment) and semantic information to improve friend recommendations in social networks; the method improves... Read more
Key finding: Designed a recommendation architecture blending content-based and collaborative filtering grounded in ontological user and item representations; employed dynamic ontology evolution to update user profiles and item... Read more