A Hybrid Word-Character Model for Abstractive Summarization
Abstractive summarization is the popular research topic nowadays. Due to the difference in language property, Chinese summarization also gains lots of attention. Most of studies use character-based representation instead of word-based to keep out the error introduced by word segmentation and OOV problem. However, we believe that word-based representation can capture the semantics of the articles more accurately. We proposed a hybrid word-character model preserves the advantage of both word-based and character-based representations. Our method also enables us to use larger word vocabulary size than anyone else. We call this new method HWC (Hybrid Word-Character). We conduct the experiments on LCSTS Chinese summarization dataset, and out-perform the current state-of-the-art by at least 8 ROUGE points.
READ FULL TEXT