Dichotomic Pattern Mining Integrated With Constraint Reasoning for Digital Behavior Analysis.

Area of Science:

• Data Mining
• Machine Learning
• Artificial Intelligence

Background:

• Sequential pattern mining faces challenges with large search spaces and redundant patterns.
• Mapping extracted patterns to specific outcomes requires further analysis.
• Existing methods struggle to effectively integrate sequential data with machine learning models.

Purpose of the Study:

• To introduce a novel pattern mining framework for semi-structured datasets.
• To leverage constraint reasoning for efficient and relevant sequential pattern extraction.
• To create pattern embeddings for improved knowledge extraction and predictive modeling.

Main Methods:

• Developed a pattern mining framework exploiting the dichotomy between outcomes.
• Employed constraint reasoning to find frequent sequential patterns with desired properties.
• Generated novel pattern embeddings from the extracted sequential patterns.

Main Results:

• The framework effectively mines sequential patterns from semi-structured data.
• Novel pattern embeddings were created, enhancing knowledge extraction and predictive modeling.
• Demonstrated utility in two real-world applications: customer intent prediction and intrusion detection.

Conclusions:

• The proposed dichotomic pattern mining framework bridges semi-structured sequential data and machine learning.
• The approach improves downstream task performance while maintaining interpretability.
• This method offers a valuable tool for analyzing complex sequential data.