GATSum: Graph-Based Topic-Aware Abstract Text Summarization

The purpose of text summarization is to compress a document into summary containing key information. abstract approaches are challenging tasks, it necessary design mechanism effectively extract salient information from the source text, and then generate summary. However, most existing difficult capture global semantics, ignoring impact on obtaining important content. To solve this problem, paper proposes Graph-Based Topic Aware Text Summarization (GTASum) framework. Specifically, GTASum seamlessly incorporates neural topic model discover potential information, which can provide document-level features for generating summaries. In addition, integrates graph network relationship between sentences through representation structure, simultaneously update local further discussion showed that latent topics help We conducted experiments two datasets, result shows superior many extractive in terms ROUGE measurement. ablation study proves has ability original subject correct improve factual accuracy summarization.