Domain specific commonsense relation extraction from bag of concepts metadata
Jiyi Li · 2015
Existing semantic knowledge bases such as WordNet and Yago contain the information of relations between entities. They do not hold the information about domain specific commonsense relations between concepts like "horse" and "farm" which intuitively have close relations on semantics in the domains of image description. Metadata which is used to describe data is widespread in the data collections of various domains and can be useful resources for relation extraction. However, keywords and tags which are important form of metadata are only list of user generated words. They do not contain syntactic information which many existing works use to extract relations. In this paper we propose an approach to collect commonsense relations for specific domains by mining knowledge of global structure and internal association in the bag of concepts from metadata of data collections. We extract commonsense relations of concepts from social tags of image datasets to show the efficiency of our solution.