Sentiment Analysis Of Weibo Topics With Sentiment Dictionary And Deep Learning

With the rapid development of the Internet,more and more people are used to expressing their views and attitudes through the Internet.Collecting and analyzing these views can bring great application value.For example,the e-commerce platform adopts the sentiment on reviews.Analysis can understand buyers 'preferences and recommend more suitable products.Social platforms can understand the public's emotional tendency to certain events through emotional analysis of the opinions of the masses,so as to control the public opinion orientation of the masses,and relevant departments can be better Monitoring of public opinion.This article uses commonly used social software Weibo as an example to conduct sentiment analysis on Weibo text.Weibo texts are different from traditional texts.Weibo texts are more concise and diverse.They are informal texts and contain a lot of abbreviations and new words.This leads to the more traditional sentiment dictionary-based sentiment analysis methods,which are largely restricted to dictionaries Perfection.In newer technologies,deep learning-based methods are greatly limited by the number of samples.When the number of samples is small,the neural network is not sufficiently trained to obtain good results.In view of the above problems,this paper proposes the idea of combining the results of two different methods to improve the effect,and simultaneously optimizes and improves the two methods.The specific work is as follows:(1)In the more traditional sentiment dictionary-based sentiment analysis,this paper proposes to expand the sentiment dictionary based on semantic similarity to improve the results of sentiment dictionary-based sentiment analysis.Use the synonym word forest to expand the affective dictionary;expand the affective dictionary based on Word2Vec;design a variety of semantic information fusion methods that can enrich the semantic vector space of words,and find related word pairs through the distance in the vector space A sentiment dictionary is expanded.These methods make up for the lack of abbreviations,new words,and spoken words in the Weibo sentiment dictionary.Through experiments,we can see that a more comprehensive sentiment dictionary effectively improves the results of sentiment analysis based on sentiment dictionary.(2)Aiming at deep learning algorithms that have performed well in various fields in recent years,a preliminary attempt was made to select a two-way gated recurrent unit neural network(BGRU)for sentiment analysis.Experimental results show that when the number of training samples is too small,the neural network model cannot Full training will lead to poor classification results.In response to this problem,we noticed that the BGRU model does not take into account the weight of each word during the training process,which ignores that each word has different influence factors on the overall sentiment of the sentence.With the attention mechanism,the BGRU-Attention model was completed by giving higher weight to important words that determine the emotion of the entire sentence.The experimental results show that the results of this model are significantly improved over the support vector machine-based model and the single-level two-way gated recurrent unit god-level network model.(3)Finally,this article attempts to combine traditional methods with emerging algorithms and use the results obtained from the sentiment dictionary to strengthen the results obtained from deep learning.Experimental results show that this method can effectively improve the performance of sentiment classification,especially For emotional categories with less corpus and insufficient training.