We present a PaperRobot who performs as an automatic research writing
assistant by (1) conducting deep and comprehensive understanding of a
large collection of existing papers in an entire target domain and
constructing a comprehensive background knowledge base (KB); (2)
creating new ideas by predicting links from the background KB, based
on a novel algorithm combining graph attention and contextual text
attention; (3) incrementally writing some key elements of a new paper
based on memory-attention networks: from the input title along with
predicted related entities to generate a paper abstract, from the
abstract to generate conclusion and future work, and finally from
future work to generate a title for a follow-on paper. Turing Tests,
where a biomedical domain expert is asked to compare the human-written
published paper and the corresponding system-generated part for the
same title, show that PaperRobot generated paper components are often
chosen over human-written ones.