Webbrain: Joint neural learning of large-scale commonsense knowledge

Abstract

Despite the emergence and growth of numerous large knowledge graphs, many basic and important facts about our everyday world are not readily available on the Web. To address this, we present Web- Brain, a new approach for harvesting commonsense knowledge that relies on joint learning from Web-scale data to fill gaps in the knowledge acquisition. We train a neural network model to learn relations based on large numbers of textual patterns found on the Web. At the same time, the model learns vector representations of general word semantics. This joint approach allows us to generalize beyond the explicitly extracted information. Experiments show that we can obtain representations of words that reflect their semantics, yet also allow us to capture conceptual relationships and commonsense knowledge.