Research And Implementation Of Natural Language Generation Based On Association Intelligence

Natural Language Generation(NLG)is an important branch in the field of Natural Language Processing(NLP),which converts input data into natural language expressions.The realization of rich and smooth natural language generation is a sign of the maturity of artificial intelligence.This research work focuses on one of the application areas of NLG technology,namely the automatic commenting task.However,in the previous works of automatic commenting based on generative models,the generated comments usually contain only information related to the article,and rarely involve association information which can often be observed in the real comments made by human.Aiming at this challenge,based on the traditional automatic commenting technology,this research work introduces association intelligence to guide the generation process of the model for the first time,making the automatically generated comments more rich and real.First of all,this research work studies how to obtain association intelligence.In this work,the association network,which has been widely studied in psychology,is used as the carrier of association intelligence.On one hand,the scale of available association network is severely restricted due to the traditional manually collecting methodology.On the other hand,the domain scope of existing association network may not match the task of automatic commenting,thus the available association network can not be used in this work.In response to this problem,an automatic word association extraction framework based on Reading Comprehension(RC)algorithm with attention mechanism is proposed in this work.The experiments are conducted on CNN and NYT datasets from which two machine association networks with about 20k association words,surpassing the existing largest human association word dataset,are finally derived.The experiments further verify that the machine association words are generally consistent with human association words with respect to semantic similarity,which highlights the promising utilization of the machine association words in the future researches of both psychology and NLP.In order to introduce the association intelligence,that is the constructed machine association network,into automatic commenting,this work proposes an automatic commenting model named Bi-Graph2Seq which is based on an encoder-decoder framework with two Graph Convolutional Network(GCN)encoders.This model converts the long content of an article which is difficult to be processed by the traditional encoder into the form of article keyword interaction graph.Dual GCN encoders are utilized to encode the keyword interaction graph and the machine association network respectively.The comments are generated by the decoder with attention mechanism and copy mechanism.The experimental results show that the proposed model can effectively integrate association intelligence into the the process of automatic comment generation,and is superior to the baseline models in terms of both article relevance and association.