Lexicon-Enhanced Transformer with Pointing for Domains Specific Generative Question Answering

Aiming at the problem of inaccurate generation caused by the lack of external knowledge in the generative automatic question answering system, we propose a new answer generation model (LEP-Transformer) that integrates domain lexicon and copy mechanism, which can enable the Transformer to effectively deal with the long-distance dependence of different text granularity and have the ability to reproduce the details of the facts when generating answers. And the experimental results on two different datasets show that the model can alleviate this problem and has ability to model short text and long text sequences simultaneously.