Research On Sentiment Analysis Algorithm Of Commodities Review Based On Convolutional Recurrent Neural Network

In recent years,with the rapid development of natural language processing technology,user sentiment analysis has become one of the key research topics in this field.It has played more and more roles in product evaluation,product design,quality monitoring,public opinion analysis,and information prediction.As an emerging natural language processing technology,the deep learning method based on convolutional recurrent neural network has been widely used in many fields such as machine translation,text classification,language recognition and text generation.The use of convolutional recurrent neural networks to perform sentiment analysis on product review text has become a very active topic in the field of sentiment analysis.At present,when the sentiment analysis model of review text based on convolutional recurrent neural network performs sentiment classification for product reviews with irregular structure,sparseness,and unclear themes,the existence of the word vector corresponding to each word is limited by the single word vector training model,which leads to neglect The problem of text keywords,while the traditional word vector is only based on the context information of the word to calculate the vector value,resulting in the inability to fully understand the semantics of the short text.Therefore,based on the structural characteristics of the short text of the review,this paper draws on the design ideas of the attention word vector,the BERT pre-training model and the convolutional recurrent neural network to study the issue of product review sentiment classification.1.Sentiment analysis method of review text based on attention word vector networkIn order to make the key information in the review text easier to obtain and improve the accuracy of the sentiment classification of the review text,the word vector corresponding to each word is limited to a single word vector training model,which leads to the problem of ignoring the text keywords.An attention-based approach is proposed.Sentiment analysis method of online comment text based on word vector.This method uses the AC-GRU network combined with the attention mechanism.The network can convert the crawled user comment text into word vectors and introduce the attention mechanism as input,and then pass through convolution and pooling,and then perform sentiment classification on the comments through the GRU network.Related experiments conducted on self-built product review data sets and public data sets show that the algorithm in this paper can improve the ability to obtain key information of review texts,and then improve the accuracy of review text sentiment classification.2.A sentiment analysis method for comment text combined with distributed crawlers and BERTIn order to obtain website product review data in time and allow word vectors to obtain richer semantic information features of sentences,a sentiment analysis method of review text combining distributed crawlers and BERT model is proposed.Aiming at the slow speed of single-machine crawlers,a distributed crawler model based on the Scrapy framework and a deduplication strategy of Bloom filters are used to crawl and store product reviews on shopping websites.Aiming at the problem that the traditional word vector only calculates the vector value based on the context information of the word,which results in the inability to fully understand the semantics of the short text,a BERT-C-GRU network combined with the BERT pre-training model is proposed.The network uses the BERT model to extract the semantic representation vector of the review text,extracts the local features of the review text through the CNN and GRU network,and then classifies the sentiment of the review text.The algorithm proposed in this paper has been experimented on a self-built user comment data set.The experimental results show that the algorithm can improve the feature representation ability of text into word vectors,thus can fully express the semantics of user comments and improve the accuracy of sentiment classification.