Mining How-to Task Knowledge From Online Communities

Nowadays, knowledge graphs have become a fundamental asset for search engines which need background commonsense knowledge for natural interactions. A fair amount of user queries seek information on problem-solving tasks such as painting a wall or repairing a bicycle. While projects like ConceptNet and Webchild have successfully compiled large amounts of knowledge on properties of objects in our daily life, there is still a big gap regarding knowledge on everyday activities, especially problem-solving tasks (how-to knowledge). Recent efforts to automatically compile commonsense have one or more the following weaknesses: (i) they ignore activity commonsense, (ii) they operate at a small scale, (iii) their outputs are not semantically organized, (iv) they are domain-specific (e.g. cooking scripts or movie scripts). All of them lack how-to knowledge. The goal of this work is to overcome these limitations and compile a large-scale, semantically organized, domain-independent formal knowledge base on tasks and task-solving steps, by tapping the contents of online communities such as WikiHow. We employ Open-IE techniques to extract noisy candidates for tasks, steps and the required tools and other items. For cleaning and properly organizing this data, we devise embedding-based clustering techniques. The resulting knowledge base, HowToKB, includes a hierarchical taxonomy of disambiguated tasks, temporal orders of sub-tasks, and attributes for involved items. A comprehensive evaluation of HowToKB shows high accuracy. As an extrinsic use case, we evaluate automatically searching related YouTube videos for HowToKB tasks.