Wednesday 8th, 09:30 - 10:10
Computer systems can understand the language that people use better if they share the background knowledge, the “common sense,” that people use when they communicate. Machine learning can be more effective and require less training data if it starts from this background of “common sense,” understanding what people are talking about in general when they use language, before learning from its input data.
We discuss the long-running open-data project ConceptNet, a multilingual knowledge graph that connects words and phrases of natural language with labeled relationships, providing background knowledge and common sense about what words mean in many languages. ConceptNet has specifically collected data in Japanese, among other languages, creating a high-quality multilingual representation that does not depend on machine translation. We then discuss how to use ConceptNet in modern machine learning and show how ConceptNet is used in a commercial setting at the enterprise feedback management company Luminoso.