Summarization of Abstractive Multi-document using Sub-graph & Network

Abstract

Automatic multi-document theoretical account system is employed to summarize many documents into a
brief one with generated new sentences. several of them are supported word-graph and ILP methodology, and much of
sentences ar unnoticed owing to the significant computation load. To cut back computation and generate decipherable
and informative summaries, we tend to propose a completely unique theoretica multi-document account system supported
chunk-graph (CG) and continual neural network language model (RNNLM). In our approach, A CG that is predicated on
word-graph is made to prepare all data during a sentence cluster, CG will scale back the scale of graph and keep a lot of
linguistics data than word-graph. we tend to use beam search and character-level RNNLM to come up with decipherable
and informative summaries from the CG for every sentence cluster, RNNLM may be a higher model to judge sentence
linguistic quality than n-gram language model. Experimental results show that our planned system outperforms all
baseline systems and reach the state-of-art systems, and also the system with CG will generate higher summaries than
that with standard word-graph.