Eugénio Rodrigues

Crowdsourced Clustering of Computer Generated Floor Plans

This paper identifies the main criteria used by architecture specialists in the task of clustering alternative floor plan designs. It shows how collective actions of respondents lead to their clustering by carrying out an online exercise. The designs were randomly pre-generated by a hybrid evolutionary algorithm and a questionnaire was posed in the end for the respondents to indicate which similarity criteria they have used. A network of designs was then obtained and it was partitioned into clusters using a modularity optimization algorithm. The results show that the main criterion used was the internal arrangement of spaces, followed by overall shape and by external openings orientation. This work allows the future development of novel algorithms for automatic classification, clustering, and database retrieval of architectural floor plans.