In this thesis, we focus on database query processing for so-called experience items, i.e., items commonly encountered in E-Commerce systems such as books, games or movies which are better described by their perceived subjective consumption experience, or Perceptual Features, than by factual meta-data normally used in SQL-style queries. To realize this, the perceived consumption experiences are extracted from social media feedback, like reviews or ratings, using methods such as Aspect-based Review Extraction or Document Embedding. These are then encoded either explicitly or implicitly as database tuples. We group similar tuples together for every item and determine the representative tuple for each of the groups, which are used in the Query-by-Example paradigm to allow users to explore and query the item space in an interactive and intuitive fashion.

In contrast to previous approaches with similar goals, like the article on ‘Exploiting Perceptual Similarity: Privacy-Preserving Cooperative Query Personalization by Lofi and Nieke’, we now introduce the notion of Shared Perspectives: paying respect to the subjectivity of user experiences. We do not try to encode only a single summarized experience for each item, but find and store dominant opinions shared by large parts of the user base instead. This allows us to represent controversial or split opinions much more accurately than previous systems, so the user can select which opinion is more relevant for them to find a similar item. We introduce the relevant conceptual foundations for Shared Perspectives, and give an overview of the design space for implementation. Furthermore, we showcase a prototype system, and evaluate it with respect to query performance as compared to previous approaches not featuring Shared Perspectives. We also investigate their semantic quality in a limited user study. As a result of these evaluations, we identify and resolve the new challenge of relevance of Perspectives, since not each commonly shared opinion is equally important or beneficial for query processing.