Designs a novel embedding model combining text embedding and graph embedding algorithms to learn the node representation for ASER, which has text as nodes and eventuality relations as edges. 

Provides useful signals for link prediction, unknown event resolution, and representation captioning on the graph, with a model utilizing BERT embeddings and LSTM.

Applied ASER to solve the commonsense reasoning problems (Section 6.2) and beat the state-of-the-art at that time.