Research And Implementation Of Short Text Sentimental Abstractive Summarization

With the rapid development of information technology and Internet technology,different kinds of information on different topics are flooding people's lives.While people can easily and quickly obtain information,the large amount of repetitive and noisy information in the network also affects the efficiency of people's access to effective information to a certain extent.Therefore,there is an urgent need for a method for effectively filtering noisy information and simplifying and consolidating useful information.At the same time,with the vigorous development of emerging media such as social network and news commentary,grasping the public's sentimental tendency toward hot events is the trend of the times,so the research on short text sentimental abstractive summarization came into being.This thesis takes the short text sentimental abstractive summarization as the research object,and combines sentiment information with traditional semantic information,in order to improve the quality of the summary and the accuracy of the sentiment expression.After in-depth research and analysis,we focus on exploring the structure of abstractive summarization models,the fusion of sentiment features and semantic features,and the optimization of sequence-to-sequence models.This thesis proposes and implements two innovative sentimental ive summarization models,which are the Sentiment-Encoder ive summarization model and the Multi-view abstractive summarization model.In order to solve the limited aspect in feature extraction of automatic summarization,the difficulty to improve sequence-to-sequence model and the inaccuracy of sentiment expression in memory network,we set several experiments on different aspects based on the Guardian Dataset.It mainly includes the influence of different network structures,parameters and memory units on the ive summarization model and the impact of the sentiment fusion algorithm on the summarization model.The experimental results show that three innovations mentioned above can effectively integrate the sentiment information into the abstractive summarization technology.The sentiment information can effectively improve the quality of summary and the accuracy of sentiment expression,and in further it can prove the effectiveness of our Short Text Sentimental Abstractive Summarization System.In addition,we also compare the abstractive summarization model with the extractive summarization model.The results show that there is still a gap between the quality of abstractive summarization and extractive summarization.More in-depth research and improvement are needed to make abstractive summarizaiont models perform better.