Experimental Comparison of Semantic Word Clouds

We study the problem of computing semantics-preserving word clouds in which semantically related words are close to each other. We implement three earlier algorithms for creating word clouds and three new ones. We define several metrics for quantitative evaluation of the resulting layouts. Then the algorithms are compared according to these metrics, using two data sets of documents from Wikipedia and research papers. We show that two of our new algorithms outperform all the others by placing many more pairs of related words so that their bounding boxes are adjacent. Moreover, this improvement is not achieved at the expense of significantly worsened measurements for the other metrics.