expert-guided cluster interpretation

Poor interpretability of clustering results—especially for complex groupings produced by black-box clustering algorithms—poses a significant challenge, as existing XAI tools struggle to generate concise, accurate, and discriminative cluster characterizations. To address this, we propose the first method that formalizes cluster explanation as a generalized Frequent Itemset Mining (gFIM) problem. Our approach employs predicate-space modeling, attribute-selection-driven optimization, and a bi-objective optimization framework balancing coverage and separation to automatically extract succinct conjunctive predicate descriptions that are both highly representative and strongly discriminative. This formulation overcomes key applicability limitations of conventional XAI techniques in clustering interpretation. Extensive evaluation across 98 benchmark datasets and a user study demonstrates that our method significantly outperforms state-of-the-art XAI baselines in both explanation quality and computational efficiency.

This work proposes an interactive, human-in-the-loop visual clustering framework for high-dimensional data, addressing the limitations of static dimensionality reduction methods that often lack interpretability and cannot incorporate human prior knowledge. By introducing a closed-loop feedback mechanism, the approach enables users to dynamically guide nonlinear projections through a small number of must-link and cannot-link constraints, while simultaneously refining low-dimensional embeddings via semi-supervised clustering. The framework further supports traceability from clustering results back to the original feature space, facilitating interpretable analysis. Experimental results on multiple benchmark datasets demonstrate that just a few rounds of user interaction significantly improve clustering quality, achieving both efficiency and interpretability in high-dimensional clustering tasks.

High-stakes domains such as medical diagnosis and financial risk management demand interpretable clustering ensembles, yet existing methods lack transparency in decision logic. Method: This paper proposes the first interpretable clustering ensemble framework, which models base clustering results as categorical variables and constructs decision trees directly in the original feature space—guided by statistical association tests (e.g., chi-square test)—to explicitly encode and trace clustering rationale. A co-designed categorical encoding scheme and ensemble strategy jointly ensure both fidelity and interpretability. Contribution/Results: The method achieves state-of-the-art clustering quality across multiple benchmark datasets while offering human-readable, auditable clustering rules. It bridges a critical gap in the literature by establishing the first principled approach to interpretable clustering ensembles, advancing both theoretical understanding and practical deployment in safety-critical applications.

This work addresses the limited semantic interpretability of clustering results in high-dimensional data after dimensionality reduction and the high expertise barrier imposed by existing visualization techniques. To bridge this gap, the authors propose an interactive framework that integrates large language models (LLMs) with visual analytics, leveraging LLMs for the first time to automatically generate human-readable semantic descriptions of clusters while incorporating external contextual knowledge. This approach substantially lowers the barrier for non-experts to understand clustering outcomes, enhancing both the accessibility and reliability of interpretability in data analysis. The effectiveness of the method is validated through systematic evaluation, and the accompanying tool has been publicly released as open-source software.

This study addresses a critical gap in clustering interpretability: existing post-hoc explanation methods primarily focus on feature importance or instance-level explanations and struggle to reliably uncover structured patterns within clusters. To systematically evaluate this limitation, the authors conduct the first controlled assessment of multiple explanation techniques—including random forest permutation importance, LIME, and principal component analysis—in synthetic datasets where ground-truth structured patterns are explicitly embedded. Results demonstrate that while these methods partially recover relevant features, none consistently identifies all types of predefined patterns. This reveals a fundamental shortcoming of current interpretability tools in capturing pattern-level cluster structure and underscores the urgent need for dedicated methods designed specifically for detecting and explaining such intra-cluster patterns.

This work addresses the redundancy problem in interpretable clustering, where distinct k-relaxed frequent patterns (k-RFPs) yield identical k-covers. The study formally characterizes, for the first time, the theoretical conditions underlying this redundancy and introduces a pattern reduction framework that retains only one representative k-RFP per unique k-cover, substantially compressing the search space. The proposed method integrates a SAT solver to generate candidate patterns, employs integer linear programming (ILP) for cluster selection, and incorporates a filtering mechanism to eliminate redundant patterns. Experimental results on multiple real-world datasets demonstrate that this strategy significantly improves computational efficiency and, in certain scenarios, further enhances clustering quality while preserving the interpretability and robustness of the selected patterns.

This study addresses the lack of systematic guidance in parameter selection and result evaluation for unsupervised data grouping methods by proposing SmartIterator, an exploratory framework grounded in a six-stage visual analytics pipeline. The approach uniquely treats the complete sequence of groupings generated through parameter sweeps as the primary analytical object, integrating quality metrics, stability assessments, member confidence scores, and domain context to deliver method-specific, actionable workflows for tasks such as clustering and topic modeling. Implemented via the IteraScope visualization system—which features semantic color encoding, group embeddings, Sankey transition flows, violin plots, and repeated prototype detection using HDBSCAN—the framework demonstrates its efficacy across three diverse datasets: social media, regional statistics, and academic publications, enabling analysts to comprehensively interpret data structures and make informed decisions.

To address the lack of a general explanation mechanism for non-interpretable clustering results, this paper proposes a generic, interpretable clustering framework based on spectral graph partitioning. The method is agnostic to specific clustering objectives and automatically fits any black-box clustering output or raw dataset into an axis-aligned decision tree, yielding structured and human-readable cluster representations. Innovatively, it introduces spectral graph partitioning—first applied to interpretable clustering—to formulate a unified graph optimization model; theoretical interpretability is established within Trevisan’s generalized framework. Moreover, several existing algorithms are unified under this graph-partitioning perspective. Experiments on multiple benchmark datasets demonstrate that the proposed method significantly outperforms mainstream baselines, achieving a superior trade-off between clustering quality and interpretability.

This work proposes an interactive ontology construction paradigm that bridges the gap between purely manual and fully automated approaches, which are often hindered by laborious processes or insufficient user control, respectively. By leveraging weighted self-organizing maps, the method enables progressive clustering of tabular data while integrating instance grouping with mechanisms for defining conceptual intensions. This approach empowers users to flexibly adjust both the number of clusters and their semantic interpretations, thereby preserving the efficiency of automation while significantly enhancing controllability. As a result, it facilitates interpretable clustering of entities and the generation of high-quality ontological classifications directly from tabular data.

This study addresses the limited reliability and diagnostic capability of existing AutoClustering systems, which stem from a lack of interpretability regarding how meta-features influence the selection of clustering algorithms and hyperparameters. For the first time, the work systematically reviews the meta-features employed across 22 AutoClustering methods and organizes them into a coherent taxonomy. By integrating global interpretability through decision predicate graphs and local interpretability via SHAP values, the authors conduct a thorough analysis of feature contributions within meta-models. Their investigation uncovers structural biases and consistent patterns in current meta-learning strategies, revealing fundamental limitations of prevailing approaches. These insights not only expose critical shortcomings but also offer actionable interpretability guidelines for designing transparent and trustworthy unsupervised AutoML systems.