The document aboutness problem asks for creating a succinct representation of a document's subject matter via keywords, sentences or entities drawn from a Knowledge Base. In this paper we propose an approach to solve this problem which improves the known solutions over all known datasets [4,19]. It is based on a wide and detailed experimental study of syntactic and semantic features drawn from the input document thanks to the use of some IR/NLP tools. To encourage and support reproducible experimental results on this task, we will make accessible our system via a public API: this is the first, and best performing, tool publicly available for the document aboutness problem.
展开▼
