A Framework for Mixed-Initiative Clustering

Mixed-initiative clustering is a machine learning task that integrates a machine’s clustering capability and a user’s guidance in order to obtain the user’s desired result. This task is different from traditional autonomous clustering tasks by introducing user criterion, a user’s understanding of data and purpose of sorting. We propose a framework for mixed-initiative clustering to solve problems such as how to model a user’s criterion and how to utilize the criterion to improve the clustering performance. Our approach consists of representing properties of the clustering model to a user and letting the user gives feedback on these properties. We also demonstrate the feasibility of this framework using an application called ”activity extraction from personal workstation contents.” Our work includes building structured activity descriptions to represent a machine’s speculation and a new clustering algorithm, the SpeClustering model, that enables extended user feedback. Furthermore, we identified the SpeClustering model as an instantiation of mixed-initiative clustering.