Prasad Shinde

Application of existing k-means algorithms for the evaluation of card sorting experiments

Card sorting is a user-centered technique that is immensely effective for designing the information architecture and the navigation structure of a website or a software application. In the card sorting experiment, the test subjects are asked to classify the cards having semantically close meaning into the same groups. The data collected from the card sorting is further analyzed to find out how frequently the cards are assigned into the same categories. There exist several cluster analysis algorithms that automate the analysis process to reduce its complexity and the required time.

The k-means algorithm is able to quickly and efficiently find the homogeneous clusters from the given data. Researchers have modified the standard k-means algorithm to overcome its limitations, so that the better clustering results can be obtained. This thesis deals with the application of the standard k-means algorithm and its modified versions on the card sorting data for the evaluation of the homogeneous clusters. In this thesis, three k-means versions are implemented by developing a software prototype. The user of the prototype can compare the clustering results produced by each k-means version, and can recommend which version produces good quality clusters and can be conveniently integrated in other existing card sorting evaluation tools.