Deep learning methods for knowledge base population
Knowledge bases store structured information about entities or concepts of the world and can be used in various applications, such as information retrieval or question answering. A major drawback of existing knowledge bases is their incompleteness. In this talk, I will present deep learning methods for automatically populating them from text, addressing the following tasks: slot filling, type-aware relation extraction and uncertainty detection. Slot filling aims at extracting information about entities from a large text corpus. I will present the system we developed for this task and especially focus on its classification module for which we designed contextCNNs, convolutional neural networks based on context splitting. In the second part of the talk, I will present our investigations of type-aware relation extraction with neural networks and introduce novel models for joint entity and relation classification: a jointly trained model and a globally normalized model. The last part of the talk focuses on assessing the factuality of statements extracted from text. I will introduce external attention, a novel attention variant which can incorporate external knowledge sources. To the best of our knowledge, we are the first to integrate an uncertainty detection component into a slot filling pipeline, extending the applicability of the system to another use case.