Causal Network Construction to Support Understanding of News

To support understanding of news, we propose a novel TEC model (Topic-Event Causal relation model) and describe the method to construct a Causal Network in the TEC model. The model includes two types of keywords to represent casual relations: topic keywords, which describe topics, and event keywords, which describe events. In the TEC model, causal relations are represented by an edge-labeled directed graph. A source vertex represents the cause of an event, and a destination vertex represents the result of that event. Each vertex contains event keywords and topic keywords and an importance score for each keyword. The edge label is the importance score of that causal relation. To construct a causal network, we extract causal relations from articles based on `clue phrases´ , merge similar event vertices and reduce the size of the causal network based on the importance score of each causal relation. Preliminary experiments to assess the validity of the proposed method demonstrated its usefulness.