Idea Generation, Creativity, and Prototypicality

We explore the use of big data tools to shed new light on the idea generation process, automatically "read" ideas to identify promising ones, and help people be more creative. The literature suggests that creativity results from the optimal balance between novelty and familiarity, which can be measured based on the combinations of words in an idea. We build semantic networks where nodes represent word stems in a particular idea generation topic, and edge weights capture the degree of novelty versus familiarity of word stem combinations (i.e., the weight of an edge that connects two word stems measures their scaled co-occurrence in the relevant language). Each idea contains a set of word stems, which form a semantic subnetwork. The edge weight distribution in that subnetwork reflects how the idea balances novelty with familiarity. Based on the "beauty in averageness" effect, we hypothesize that ideas with semantic subnetworks that have a more prototypical edge weight distribution are judged as more creative. We show this effect in eight studies involving over 4,000 ideas across multiple domains. Practically, we demonstrate how our research can be used to automatically identify promising ideas and recommend words to users on the fly to help them improve their ideas.