Relevance- and interface-driven clustering for visual information retrieval
作者：
Highlights：
• We present a novel relevance-driven clustering algorithm for Visual IR.
• We present expected F1-score (EF1) as a new objective for clustering in IR.
• We demonstrate that the optimal solution to EF1 maximization can be cast as a MILP.
• We present two efficient greedy algorithms for optimizing EF1.
• Experiments show that relevance-driven clustering improves user performance.
• We provide the source code as an open-source library.
摘要
•We present a novel relevance-driven clustering algorithm for Visual IR.•We present expected F1-score (EF1) as a new objective for clustering in IR.•We demonstrate that the optimal solution to EF1 maximization can be cast as a MILP.•We present two efficient greedy algorithms for optimizing EF1.•Experiments show that relevance-driven clustering improves user performance.•We provide the source code as an open-source library.
论文关键词：Visual information retrieval,Relevance-driven Clustering,Visual search user study,Clustering via filter optimization
论文评审过程：Received 29 April 2019, Revised 27 April 2020, Accepted 8 July 2020, Available online 13 July 2020, Version of Record 16 July 2020.
论文官网地址：https://doi.org/10.1016/j.is.2020.101592
