Joint Learning-based Heterogeneous Graph Attention Network for Timeline Summarization

Timeline summarization (TLS) is defined as a task for summarizing events in chronological order, which gives readers comprehensive understanding of an evolutionary story. Previous studies on the timeline ignored information interaction between sentences and dates, adopted pre-defined unlearnable representations them, significantly degrade performance. They also considered date selection event detection two independent tasks, makes it impossible to integrate their advantages obtain globally optimal summary. In this paper, we present {joint learning-based heterogeneous graph attention network TLS (HeterTls), are combined into unified framework improve extraction accuracy remove redundant simultaneously. Our involves multiple types nodes, iteratively learned across layer. We evaluated our model four datasets, found that outperformed current state-of-the-art baselines with regard ROUGE scores metrics.