LayerFolding: Discovering Creative Links in Word Association Networks

A frequent challenge in creative tasks such as advertising is finding novel and concrete representations of abstract concepts. We cast this problem as finding, in word association networks, the relevant indirect associations of a given node. We propose a novel approach, LayerFolding, which selects nodes at increasing distances from the given node, according to their relatedness to it. The relatedness is calculated based on the shortest paths that are potentially coherent. In a test against a small set of visual representations of abstract concepts found in real advertisements, LayerFolding provides a 79\% recall, and outperforms other two popular semantic relatedness measures.