Knowledge Base Inference using Bridging Entities

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2038–2043, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Knowledge Base Inference using Bridging Entities Bhushan Kotnis Indian Institute of Science bkotnis@dese.iisc.ernet.in Pradeep Bansal Indian Institute of Science pradeepb@ee.iisc.ernet.in Partha Talukdar Indian Institute of Science ppt@serc.iisc.in Abstract Large-scale Knowledge Bases (such as NELL, Yago, Freebase, etc.) are often sparse, i.e., a large number of valid rela- tions between existing entities are missing. Recent research have addressed this prob- lem by augmenting the KB graph with ad- ditional edges mined from a large text cor- pus while keeping the set of nodes fixed, and then using the Path Ranking Algo- rithm (PRA) to perform KB inference over this augmented graph. In this paper, we extend this line of work by augmenting the KB graph not only with edges, but also with bridging entities, where both the edges and bridging entities are mined from a 500 million web text corpus. Through experiments on real-world datasets, we demonstrate the value of bridging entities in improving the performance and running time of PRA in the KB inference task. 1 Introduction Large-scale knowledge bases (KB) like Freebase (Bollacker et al., 2008), Yago (Suchanek et al., 2007), NELL (Mitchell et al., 2015) can be use- ful in a variety of applications like natural lan- guage question answering, semantic search en- gines, etc. These knowledge bases consist of mil- lions of real world entities and relationships be- tween them which are stored in the form of a di- rected graph where links represent relations and nodes represent the entities. Although such KBs contain millions of entities, they are still very sparse, i.e., they are missing a large number of relations between existing entities (West et al., 2014). Performing inference over the knowledge graph for predicting relations between two entities is one way of densifying the KB graph. For example, Figure 1: Example showing how addition of the bridging entity, Brian McCain, and the two edges incident on it can help the PRA algorithm (Lao and Cohen, 2010) to infer the initially missing relation instance teamPlaysSport(Yankees, BaseBall). The original KB graph consisted only of two nodes, Yankees and Baseball, and no edges. from (Germany, playsinTournament, FIFA) and (FIFA, tournamentofSport, Soccer), we can infer (Germany, playsSport, Soccer). The Path Ranking Algorithm (PRA) (Lao and Cohen, 2010), (Lao et al., 2011) performs such an inference by learning inference rules over the knowledge graph. If the knowledge graph is sparse, i.e., if there are a very few or no paths between source and target entities, then PRA is unable to predict the existence of a relation. To address this shortcom- ing, (Lao et al., 2012) augmented the knowledge graph with paths obtained from an external corpus. The added paths consisted of unlexicalized depen- dency labels obtained from a dependency parsed external corpus. To improve the expressivity of the added paths, instead of the unlexicalized la- bels, (Gardner et al., 2013) augmented the KB graph with verbs …

Content truncated to 500 words