Data-driven digital memory applications lack predefined navigation paths and strict hierarchical structures. They are based on large collections of memory items that can become overwhelming to users. Recommender systems can improve user experience through the proposal of personalized relevant items. However, very little academic literature has been dedicated to discussing this type of filtering of digital memory resources and the provision of customized contents to active users. In this paper, an architecture of a hybrid enhanced recommender (HER) system, which integrates collaborative filtering and content based filtering techniques and resolves most of the weaknesses of the individual approaches. This architecture also proposes an ontology to build semantic user profiles and represent memory items to mitigate the lack of semantics of traditional content-based method. Through combining those techniques, this architecture has the potential to cope with data sparsity problems, avoid overspecialization issues and partially resolve cold start problems.
Paper
Full text
THE APPLICATION OF RECOMMENDER SYSTEMS TO DATA-DRIVEN DIGITAL MEMORY
Semantic Scholar · Computer Science · 2019
Abstract
Data-driven digital memory applications lack predefined navigation paths and strict hierarchical structures. They are based on large collections of memory items that can become overwhelming to users. Recommender systems can improve user experience through the proposal of personalized relevant items. However, very little academic literature has been dedicated to discussing this type of filtering of digital memory resources and the provision of customized contents to active users. In this paper, an architecture of a hybrid enhanced recommender (HER) system, which integrates collaborative filtering and content based filtering techniques and resolves most of the weaknesses of the individual approaches. This architecture also proposes an ontology to build semantic user profiles and represent memory items to mitigate the lack of semantics of traditional content-based method. Through combining those techniques, this architecture has the potential to cope with data sparsity problems, avoid overspecialization issues and partially resolve cold start problems.