Abstract
Summarization of news articles is becoming crucial for allowing quick and concise access to information about daily events. This problem has been intensively studied in the literature through extractive and abstractive summarization techniques. However, these techniques are static and thus do not satisfy user needs in having summaries with specific structures or details. Similarly, query-based summarization techniques fail to handle content-independent queries that target the type of summary information such as time, location, reasons, and consequences of reported events.
The NEWS system presented in this demo supports multigranular summarization along two dimensions: the level of detail and type of information. The system employs a finegrained information extraction strategy able to extract facts and their related facets with type tagging. The extracted information is then modeled as a graph used to create summaries. NEWS provides an algorithm that incrementally expands summaries based on the nodes visited by users folding related events into the search space. The demonstration uses a set of news articles from DUC and TAC datasets.