JOLTER: Joint Learning of Topics, Entities, Events and Relations

In Phase I of JOLTER, we will demonstrate how unsupervised models for discovering entities, events, topics, and relations allow for a domain independent analysis of technical literature.We will demonstrate (1) how to discover entities, events, and their classes in a fully unsupervised manner; (2) how generative models not only provide a flexible platform for the automatic discovery of entities, events, and their relations, but also enable human interpretation by allowing minor adjustments (such as changing a topic or entity distribution) and observing the effect on the other variables; (3) how individual models that have been shown to be successful for discovering categories and relations can be combined to form a joint model; (4) how to evaluate such models through automatic and human-centric methods.