A Compare-Aggregate Model with External Knowledge for Query-Focused Summarization

Query-focused extractive summarization aims to create a summary by selecting sentences from original document according to query relevance and redundancy. With recent advances of neural network models in natural language processing, attention mechanism is widely used to address text summarization task. However, existing methods are always based on a coarse-grained sentence-level attention, which likely to miss the intent of query and cause relatedness misalignment. To address the above problem, we introduce a fine-grained and interactive word-by-word attention to the query-focused extractive summarization system. In that way, we capture the real intent of query. We utilize a Compare-Aggregate model to implement the idea, and simulate the interactively attentive reading and thinking of human behavior. We also leverage external conceptual knowledge to enrich the model and fill the expression gap between query and document. In order to evaluate our method, we conduct experiments on DUC 2005–2007 query-focused summarization benchmark datasets. Experimental results demonstrate that our proposed approach achieves better performance than state-of-the-art.