Title Generation with Recurrent Neural Network

Abstract

We proposed a title generation method with a recurrent neural network using concepts of machine translation. The title generator consists of an encoder and a decoder and they are constructed with Long Short Term Memory, which is one of the recurrent neural networks. We construct a distributed representation of an article in the encoder and the decoder generates a title without extraction of an article. In some evaluational experiments we confirmed that our proposed method could generate appropriate titles from articles but in some articles the method generate random titles.