Personalized top-k skyline queries in high-dimensional space

As data of an unprecedented scale are becoming accessible, it becomes more and more important to help each user identify the ideal results of a manageable size. As such a mechanism, skyline queries have recently attracted a lot of attention for its intuitive query formulation. This intuitiveness, however, has a side effect of retrieving too many results, especially for high-dimensional data. This paper is to support personalized skyline queries as identifying “truly interesting” objects based on user-specific preference and retrieval size k. In particular, we abstract personalized skyline ranking as a dynamic search over skyline subspaces guided by user-specific preference. We then develop a novel algorithm navigating on a compressed structure itself, to reduce the storage overhead. Furthermore, we also develop novel techniques to interleave cube construction with navigation for some scenarios without a priori structure. Finally, we extend the proposed techniques for user-specific preferences including equivalence preference. Our extensive evaluation results validate the effectiveness and efficiency of the proposed algorithms on both real-life and synthetic data.     

Knowledge distillation meets recommendation: collaborative distillation for top-N recommendation

Knowledge and Information Systems - Knowledge distillation (KD) is a successful method for transferring knowledge from one model (i.e., teacher model) to another model (i.e., student model)....     

Toward efficient multidimensional subspace skyline computation

Skyline queries have attracted considerable attention to assist multicriteria analysis of large-scale datasets. In this paper, we focus on multidimensional subspace skyline computation that has been actively studied for two approaches. First, to narrow down a full-space skyline, users may consider multiple subspace skylines reflecting their interest. For this purpose, we tackle the concept of a skycube, which consists of all possible non-empty subspace skylines in a given full space. Second, to understand diverse semantics of subspace skylines, we address skyline groups in which a skyline point (or a set of skyline points) is annotated with decisive subspaces. Our primary contributions are to identify common building blocks of the two approaches and to develop orthogonal optimization principles that benefit both approaches. Our experimental results show the efficiency of proposed algorithms by comparing them with state-of-the-art algorithms in both synthetic and real-life datasets.     

The scientific impact and partner selection in collaborative research at Korean universities

This study seeks to bridge the gap between scientometrics literature on scientific collaboration and science and technology management literature on partner selection by linking scientists’ collaborator preferences to the marginal advantage in citation impact. The 1981–2010 South Korea NCR (National Citation Report), a subset of the Web of Science that includes 297,658 scholarly articles, was used for this research. We found that, during this period, multi-author scientific articles increasingly dominated single-author articles: multi-university collaboration grew significantly; and the numbers of research publications produced by teams working within a single institution or by a single author diminished. This study also demonstrated that multi-university collaboration produces higher-impact articles when it includes “Research Universities,” that is, top-tier university schools. We also found that elite universities experienced impact degradation of their scientific results when they collaborated with lower-tier institutions, whereas their lower-tier partners gained impact benefits from the collaboration. Finally, our research revealed that Korean universities are unlikely to work with other universities in the same tier. This propensity for cross-tier collaboration can be interpreted as strategic partner selection by lower-tier schools seeking marginal advantage in citation impact.     

Hybrid entity clustering using crowds and data

Query result clustering has attracted considerable attention as a means of providing users with a concise overview of results. However, little research effort has been devoted to organizing the query results for entities which refer to real-world concepts, e.g., people, products, and locations. Entity-level result clustering is more challenging because diverse similarity notions between entities need to be supported in heterogeneous domains, e.g., image resolution is an important feature for cameras, but not for fruits. To address this challenge, we propose a hybrid relationship clustering algorithm, called Hydra, using co-occurrence and numeric features. Algorithm Hydra captures diverse user perceptions from co-occurrence and disambiguates different senses using feature-based similarity. In addition, we extend Hydra into ${\mathsf{Hydra }_\mathsf{gData }}$ Hydra gData with different sources, i.e., entity types and crowdsourcing. Experimental results show that the proposed algorithms achieve effectiveness and efficiency in real-life and synthetic datasets.