General purpose computer-assisted clustering and conceptualization.

Area of Science:

• Data science
• Computer science
• Bioinformatics

Background:

• Numerous automated cluster analysis methods exist, each optimizing different objective functions.
• Determining the most insightful or useful clustering for a specific user and dataset is challenging.
• Existing methods lack user-centric exploration for discovering meaningful data partitions.

Purpose of the Study:

• To develop a computer-assisted method for discovering insightful conceptualizations through data clustering.
• To enable users to explore a metric space of partitions for enhanced data understanding.
• To facilitate the discovery of useful information beyond fully automated approaches.

Main Methods:

• Developed a metric space encompassing existing and novel clusterings of input objects.
• Integrated millions of additional solutions based on combinations of existing clusterings.
• Enabled user interaction and exploration within the metric space of partitions.

Main Results:

• Demonstrated that the computer-assisted approach facilitates more efficient and insightful discovery of useful information.
• Showcased superior performance compared to expert human coders and existing fully automated methods.
• Implemented evaluation designs to assess the approach's performance on specific data types.

Conclusions:

• The developed method significantly enhances the discovery of insightful data conceptualizations.
• User-assisted exploration of a diverse partition space leads to more effective data analysis.
• This approach offers a powerful alternative to traditional automated clustering and manual analysis.