580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can adaptive tree-based data structures enable efficient interactive mining under varying minimum support thresholds?
This research area focuses on developing specialized tree data structures that separate the mining model construction from mining execution, enabling interactive mining where users can adjust the minimum support threshold (minsup) without needing to rescan or rebuild the entire database. This is crucial for improving responsiveness and reducing computational overhead in real-world data mining scenarios where iterative exploratory analysis is common.
Key finding: Proposes a two-layer model distinguishing mining model construction and mining process layers that enables efficient interactive frequent pattern mining by avoiding rescanning and reconstructing the mining model when minsup... Read more
Key finding: Introduces two novel tree structures (IWFPTWA and IWFPTFD) and algorithms that enable single-pass incremental and interactive weighted frequent pattern mining, accommodating updates and minsup changes efficiently without... Read more
Key finding: Extends granular computing by modeling computations on complex granules controlled interactively by agents, supporting dynamic, adaptive, and interactive computations. This abstract notion underpins adaptive data structures... Read more
Key finding: Proposes the IDFP-tree, a compact and complete tree structure that decouples mining model construction from mining process for frequent subgraph pattern mining. It supports interactive mining by enabling multiple minsup... Read more
Key finding: Extends the FP-tree concept for uncertain data mining with UDFP-tree supporting interactive mining under uncertainty. It allows efficient updates and resizing under changing minsup without reconstructing mining models,... Read more
2. What techniques improve user-driven iterative exploration and recommendation in interactive database mining to overcome slow convergence?
This theme investigates systems and algorithmic frameworks to support users in interactive data exploration by recommending queries, incorporating active learning, and exploiting database query properties to overcome slow convergence in user modeling. The goal is to enable rapid, interpretable insights with minimal user effort in complex and high-dimensional databases.
Key finding: Develops active learning techniques integrated with database query semantics to accelerate convergence in explore-by-example frameworks, reducing labeled examples needed from hundreds to manageable quantities by leveraging... Read more
Key finding: Introduces a query recommendation system leveraging fragmentation and comparison strategies to suggest relevant queries based on past user queries, improving the efficiency and guidance of interactive database exploration by... Read more
Key finding: Presents RAPID, a scalable system for real-time interactive data mining on Twitter data, featuring dynamic query expansion and interactive visual analytics that enable users to iteratively refine and control data collection... Read more
Key finding: Proposes an incremental association rule mining algorithm tailored for dynamically changing web logs, enabling efficient interactive mining that avoids redundant database scans and candidate generation, facilitating... Read more
Key finding: Develops the Generator-Enumeration Tree (GE-tree) data structure and associated algorithms (PSM+ and PSM-) for incremental and decremental maintenance of frequent pattern spaces, enabling fast interactive mining and real-time... Read more
3. How can interactive and user-in-the-loop clustering and knowledge discovery improve interpretability and adaptability in complex data mining tasks?
Research under this theme explores techniques incorporating human interaction and feedback into clustering and knowledge discovery processes, blending computational power with user expertise to generate more meaningful, interpretable, and adaptable models. This includes frameworks for interactive clustering, user-guided synthesis of process models, and approaches bridging analysis and symbolic knowledge construction.
Key finding: Comprehensively surveys interactive clustering methods emphasizing user involvement at different stages, types of operations, feedback incorporation, and evaluation metrics. It highlights how user interaction significantly... Read more
Key finding: Proposes a modified α-algorithm for Petri net discovery that incrementally involves domain experts during the mining process, employing intermediate data and human knowledge to iteratively converge to more accurate and... Read more
Key finding: Discusses the philosophical foundations differentiating algorithmic computation from interactive computation, arguing the empirical power of interaction to model dynamic, real-world processes beyond traditional Turing machine... Read more
Key finding: Argues for combining human perceptual strengths and machine computational power in interactive visual mining systems. Demonstrates how integrating visualisation, genetic algorithm-based rule induction, and user feedback... Read more
Key finding: Describes DAMIS, a web-based cloud-enabled scientific workflow tool with drag-and-drop interfaces facilitating interactive modelling, simulation, and understanding of complex data mining processes. Enables combined use of... Read more