Commonsense knowledge acquisition is a long-standing goal of AI. Recent
efforts to automatically compile commonsense either: (i) ignore activity
commonsense, (ii) operate at a small-scale, (iii) are not semantically
organized, (iv) are domain specific (e.g. over cooking scripts or movie
scripts). The goal of this work is to overcome these limitations and
compile a large-scale, semantically organized, domain independent activity
commonsense knowledge base.
An interesting input source of activity knowledge is how-to forums such as
WikiHow.com – containing rich textual and visual descriptions on common
activities. To answer a how-to question, for instance, “How to paint a
wall?”, Wikihow provides for us a list of activities, step by step, along
with useful information like images, things required, etc. However, it is
not organized for machine reading and still far from a semantically
organized knowledge base. We leverage the semi-structured data from
Wikihow, using open information extraction methods and by proposing a
method to organize this knowledge.