Automating News Summarization with Sentence Vectors Offset

Text summaries consist of short versions of texts that convey their key aspects and help readers understand the gist of such texts without reading them in full. Generating such summaries is important for users who must sift through ever-increasing volumes of the content generated on the web. However, generating high-quality summaries is time-consuming for humans and challenging for automated systems, since it involves understanding the semantics of the underlying texts in order to extract key information. In this work, we develop an extractive text summarization method using vector offsets, which we show empirically to be able to summarize texts from an Internet news corpus with an effectiveness competitive with state-of-the-art extractive techniques.