Researches On Sentiment Analysis Of Network Comments Based On Deep Learning

With the development of the network,many people express their opinions and comments about things on the Internet.Through the analysis of these data,we can obtain different people's emotional tendencies towards certain things.How to analyze online comment texts is a hot topic in natural language processing tasks,and sentiment analysis is also an important task.While subjective and objective analysis is particularly important to achieve the integrated design for emotional classification.In this paper,the optimization of the sentiment analysis algorithm and the subjective and objective analysis algorithm are deeply studied.The main work is as follows:This paper designs and implements a sentiment analysis algorithm GCNN-POS(Gated Convolutional Neural Networks based on Part-of-Speech Tagging)based on convolutional neural networks and recurrent neural networks.First of all,the same word has different parts of speech(POS)and meaning in different scenes.Based on the corpus after word segmentation and tagged,the algorithm uses the word2 vec method to train the corpus,transforms the corpus into word vectors based on part-of-speech tagging.Then,the algorithm utilizes the bidirectional gated recurrent unity(BiGRU)neural network layer to extract the semantic information about the context,and further uses the convolutional neural networks(CNN)to extract features of texts and lower the dimension of word vectors.Finally,the sigmoid function is used to convert the results of the experimental scores into corresponding probabilities for classification.For the first time,the deep learning algorithm is applied to the subjective and objective analysis tasks,and the AT-BiGRU(Bidirectional Gated Recurrent Units based on Attention Mechanism)algorithm based on the subjective and objective analysis tasks is proposed.This paper no longer relies on the method of feature engineering,but the better performing deep learning methods to learn the characters of subjective and objective texts.We add the attention mechanism on the bidirectional gated recurrent unity neural network to allocate the weight of deep information,and the important word weights and positions in the texts can be calculated.In order to verify the validity of the proposed algorithm,this paper conducts a comparative experiment of sentiment analysis on ChnSentiCorp and IMDB corpus.Experimental results show that the GCNN-POS model achieves state-of-the-art performance on the sentiment analysis tasks,with 95.1% and 90.3% F1-score on ChnSentiCorp corpus and IMDB corpus respectively.We conduct experiments on subjective and objective corpus of Weibo comments about deep learning method AT-BiGRU.Compared with traditional methods like the statistical method and the traditional machine learning-based methods,the method we propose reduces the process of manually extracting features and simplifies the operation.This method we propose achieves a good classification F1-score of 76.2%.