50 Ways to Bake a Cookie: Mapping the Landscape of Procedural Texts

However, as content However, a single task might have thousands of corresponding is created independently, a single task could have thousands of procedural texts. This is both due to variations (for example, different corresponding procedural texts. This makes it difficult for users to recipes for the same dish) and due to the distributed nature view the bigger picture and understand the multiple ways the task of the web, where content is created independently by people who could be accomplished. In this work we propose an unsupervised do not communicate. Thus, looking at one (or a few) procedural learning approach for summarizing multiple procedural texts texts only gives the reader a limited view of the possibilities. Consequently, into an intuitive graph representation, allowing users to easily explore when people try to determine the best choice for them commonalities and differences. We demonstrate our approach given preferences (e.g., taste) and constraints (budget, time, items on recipes, a prominent example of procedural texts. User studies they do or do not have), they often resort to extensive browsing show that our representation is intuitive and coherent and that and comparisons between different texts to get the bigger picture.