Text Sentiment Analysis And Research For User Reviews

In recent years,along with the Internet is widely used in all walks of life,the Internet based businesses in today's information age is high-speed development,the representative enterprises include alibaba,jingdong,weibo,twitter,merchants eager need to understand the user to product emotional attitude and the demand for products,In order to make the corresponding improvement to the product service and product quality of the business,improve the efficiency.At the same time,"We Media" is also a key way for individuals to express their views.With the continuous increase of netizens' attitudes or emotions towards products and services,a large number of comments containing personal views have been accumulated.These online comments are not only large in number and varied in style,but also unstructured text data.It is of great significance to extract emotional information from online review data,judge people's judgment attitude towards products and services,and determine the focus and emotional tendency of users' comments.In this context,this paper takes user comments as the research object,and studies and analyzes two sub-tasks of text sentiment classification,namely text vector representation and feature extraction.First of all,there is redundancy and noise in the processing of long text sequence data,so that the current sentiment classification algorithm is insufficient to extract the long text sequence information in the feature extraction stage,and the classification accuracy is not accurate.To solve these problems,this paper proposes an improved BERT algorithm model based on text filter.The model filter passes through the text in the original data set to build a new data set,and then through the model filter out text review data in coarse-grained aspects related statements,after the above steps to complete the sentence is adopted to form as the improved BERT model input,and in the process of the output to join attention mechanism,The experimental results show that the proposed model can improve the performance of the aspect-level text sentiment analysis algorithm in processing long text tasks to a certain extent.Secondly,in order to solve the problem that the classification accuracy is not high due to only paying attention to the whole emotion while ignoring the specific details in the process of text sentiment classification,this paper proposes an improved Bert algorithm combining text filtering with the aspect level text sentiment classification model of bidirectional gating loop unit.In this hierarchical model,the word segmentation obtained after data preprocessing is firstly transformed into a sentence-level low-dimensional and dense text vector representation,and then the sentence-level text vector representation of the previous step is input into the convolutional neural network and bidirectional gating unit for feature extraction.Then,in the decoding stage of the output of discourse level semantic information,user features and product features are added to obtain the final discourse level semantic information,and the classification function is used to classify the final discourse level semantic information.Finally,in order to verify the effectiveness of the overall model,several experiments were carried out on the proposed model.In the experiments,the overall modeling model was compared with the related classification model,and it was proved that the performance of the proposed method was improved compared with other related network models.