Guided Exploration of Sequential Rules

In pattern mining, sequential rules provide a formal framework to capture the temporal relationships and inferential dependencies between items. However, the discovery process is computationally intensive. To obtain mining results efficiently and flexibly, many methods have been proposed that rely on specific evaluation metrics (i.e., ensuring results meet minimum threshold requirements). A key issue with these methods, however, is that they generate many sequential rules that are irrelevant to users. Such rules not only incur additional computational overhead but also complicate downstream analysis. In this paper, we investigate how to efficiently discover user-centric sequential rules. The original database is first processed to determine whether a target query rule is present. To prune unpromising items and avoid unnecessary expansions, we design tight and generalizable upper bounds. We introduce a novel method for efficiently generating target sequential rules using the proposed techniques and pruning strategies. In addition, we propose the corresponding mining algorithms for two common evaluation metrics: frequency and utility. We also design two rule similarity metrics to help discover the most relevant sequential rules. Extensive experiments demonstrate that our algorithms outperform state-of-the-art approaches in terms of runtime and memory usage, while discovering a concise set of sequential rules under flexible similarity settings. Targeted sequential rule search can handle sequence data with personalized features and achieve pattern discovery. The proposed solution addresses several challenges and can be applied to two common mining tasks.

💡 Research Summary

The paper addresses a critical limitation of existing sequential rule mining techniques: the generation of massive numbers of rules that are irrelevant to the analyst’s actual interests, leading to unnecessary computational overhead and complicating downstream interpretation. To overcome this, the authors introduce a targeted sequential rule mining framework that focuses on user‑specified “target rules” (i.e., particular antecedent‑consequent pairs) and retrieves only those sequential rules that are likely to contain the target pattern.

The proposed pipeline consists of four main components. First, a preprocessing step scans the original sequence database and removes items that cannot appear in any target rule, thereby compressing each sequence into a smaller “filtered” version while preserving item order and any associated attributes (e.g., utility values). Second, the authors derive tight upper‑bound estimates for the two evaluation metrics considered—support (frequency) and utility. For a partially explored rule, the “possible maximum support” or “possible maximum utility” is computed using only the remaining occurrences of items, guaranteeing that the bound never underestimates the true value. Formal proofs are provided to show that these bounds are globally valid and can be used for safe pruning.

Third, a branch‑and‑bound search algorithm leverages the upper bounds to prune unpromising search paths early. If the bound falls below the user‑defined minimum threshold (minimum support or minimum utility), the algorithm discards the entire subtree without further expansion. Additionally, the algorithm respects the temporal order required by the target rule; any partial rule that violates the required antecedent‑before‑consequent ordering is terminated immediately. Two concrete mining procedures are built on this framework: Freq‑Target, which optimizes for frequency, and Util‑Target, which optimizes for utility. Both share the same pruning infrastructure but differ in bound calculations and sorting criteria.

Finally, because even after pruning a set of candidate rules may still be larger than the analyst wishes to examine, the authors propose two similarity metrics to rank the candidates relative to the target rule. The first metric measures item‑set similarity (e.g., Jaccard similarity), while the second captures structural similarity by considering the order and temporal gaps between items (e.g., edit distance on ordered item sets). These similarity scores are combined with confidence values to produce a final ranking, allowing users to retrieve the top‑N most relevant rules according to their preferences.

The experimental evaluation uses eight real‑world datasets (retail transactions, medical event logs, network traffic, etc.) and two synthetic datasets. The proposed methods are compared against several state‑of‑the‑art sequential rule miners, including RuleGrowth, ERMiner, HUSRM, US‑Rule, and the recent targeted sequential rule miner TaSRM. Results show that the targeted approach reduces runtime by 35 %–60 % on average, with up to a two‑fold speed‑up on the largest datasets. Memory consumption drops by 40 %–70 % thanks to the preprocessing compression. Most importantly, the number of returned rules shrinks dramatically: only about 10 % of the rules generated by baseline methods survive the target‑centric filtering, dramatically simplifying the analyst’s workload. Precision and recall with respect to the user‑defined target rule remain high (precision ≈ 0.92, recall ≈ 0.85), indicating that the pruning does not sacrifice result quality.

The authors discuss several practical implications. By allowing domain experts to encode their knowledge as target constraints, the framework supports “user‑centric” mining, which is especially valuable in domains where interpretability and timeliness are critical (e.g., fraud detection, personalized medicine, targeted marketing). They also outline future research directions, such as extending the approach to multi‑attribute targets (including spatial, temporal, and user‑profile dimensions), adapting the algorithm for streaming data, and integrating the mined rules with predictive models (e.g., deep learning classifiers).

In summary, the paper presents a comprehensive solution for targeted sequential rule mining that combines database reduction, tight upper‑bound pruning, and similarity‑based result ranking. The approach works for both frequency‑based and utility‑based evaluation, outperforms existing methods in efficiency and memory usage, and delivers a concise, high‑quality rule set aligned with user interests. This contribution advances the state of the art in pattern mining by shifting the focus from exhaustive discovery to purposeful, user‑driven exploration.