Towards Semantically AwareWord Cloud Shape Generation

Word clouds are a data visualization technique that showcases a subset of words from a body of text in a cluster form, where a word's font size encodes some measure of its relative importancetypically frequency-in the text. This technique is primarily used to help viewers glean the most pertinent information from long text documents and to compare and contrast diferent pieces of text. Despite their popularity, previous research has shown that word cloud designs are often not optimally suited for analytical tasks such as summarization or topic understanding. We propose a solution for generating more efective visualization technique that shapes the word cloud to refect the key topic(s) of the text. Our method automates the processes of manual image selection and masking required from current word cloud tools to generate shaped word clouds, better allowing for quick summarization. We showcase two approaches using classical and state-of-the-art methods. Upon successfully generating semantically shaped word clouds using both methods, we performed preliminary evaluations with 5 participants. We found that although most participants preferred shaped word clouds over regular ones, the shape can be distracting and detrimental to information extraction if it is not directly relevant to the text or contains graphical imperfections. Our work has implications on future semantically-aware word cloud generation tools as well as eforts to balance visual appeal of word clouds with their efectiveness in textual comprehension.