Human-supervised clustering of multidimensional data using crowdsourcing.

Area of Science:

• Data Science
• Human-Computer Interaction
• Machine Learning

Background:

• Clustering is vital in data analysis, but lacks universal metrics, especially for high-dimensional data.
• Human perception is challenged by high-dimensional data, complicating cluster identification.
• Expert consensus is often required, which is difficult to scale.

Purpose of the Study:

• To design a mobile human-computing game to leverage human perception for multidimensional data clustering.
• To develop and evaluate clustering algorithms that integrate aggregated human answers.
• To explore the efficacy of crowdsourcing for abstract computational problems.

Main Methods:

• Development of a mobile game to collect human judgments on data clusters.
• Proposal of two clustering algorithms utilizing aggregated human responses.
• Experimental validation on synthetic and real-world high-dimensional datasets.

Main Results:

• Proposed hybrid clustering methods perform comparably to or better than leading automated algorithms.
• Human-computer interaction games effectively query perceptual insights for clustering.
• Crowdsourced annotations provide a viable strategy for collective intelligence in data analysis.

Conclusions:

• Hybrid systems combining automated methods with crowdsourced human perception are effective for complex clustering tasks.
• Mobile gaming platforms can be utilized as tools for scientific data analysis.
• Leveraging collective human wisdom through crowdsourcing offers a scalable solution for challenging computational problems.