Deep Learning Natural Language Processing System For Network Public Opinion Analysis

With the rapid development of the Internet,the explosive growth of online social information has also brought the problem of online public opinion analysis.The traditional network public opinion analysis mode uses the method of thesaurus,and the corpus is directly compared with the thesaurus for judgment.Due to the complexity of Chinese,such as the presence of non-standard Chinese expressions such as near-tone words,synonyms,acronyms,and cryptic words,the effect of public opinion analysis is poor.Combining deep learning to process the corpus can effectively improve the accuracy of the results when analyzing non-standard Chinese expressions.Based on this method,this paper deeply researches natural language processing based on deep learning,and applies deep learning to natural language processing in order to obtain more accurate results in analyzing the similarity of words in natural language,and combines this method to develop Deep learning natural language processing system for network public opinion analysis.The research content of this article mainly includes the following parts.Research on Scrapy web crawler based on python.The article will use web crawlers to obtain real-time corpus data from the network to ensure the timeliness of the database.In this way,the coverage of natural language processing on non-standard languages can be effectively improved,and the analysis effect can be improved;Build a corpus database server.This paper needs to constantly update the existing corpus in the process of processing data,so it is necessary to build a corresponding corpus data server to store real-time corpus data and complete the preliminary processing of corpus data in the database.Segmentation to get data that can be used for deep learning;Design and Implementation of Deep Learning Algorithms for Natural Language Processing Based on TensorFlow.This article adopts a method of word similarity analysis based on the fusion of dynamic weights and multiple models.It selects different corpora according to the characteristics of the corpus and combines multiple models to perform calculations to improve the accuracy of word similarity analysis.The obtained results have better support for online public opinion monitoring.In this article,it is found through experiments that the multi-model fusion method has better results than the single model.The PKU-500 dataset was used in the NLPCC-ICCPOL 2016 Chinese Word Similarity Competition.As a reference standard for evaluation,the word similarity analysis method of the dynamic weight multi-model fusion adopted in this system obtained a Spearman rank correlation coefficient of 0.568,which was an increase of 9.6% compared to the result of the first place in the competition,so the multi-model The fusion method can improve the accuracy when calculating word similarity;Integrate the above parts to build a network public opinion analysis system.The built network public opinion analysis system will automatically collect real-time network corpora and join corpora for deep learning calculations,continuously update the calculation results,improve the timeliness of the network public opinion analysis system,and provide word similarity query functions.Users can use this system Get the quantitative result of the similarity of two words directly.