Sentiment Analysis Of Review Texts Based On Deep Learning

Sentiment analysis is a basic task in natural language processing.In sentiment analysis tasks,document-level sentiment analysis is a more common way.At present,various modern technologies are constantly updated,and a large number of user comments appear on major e-commerce platforms,which usually reflect users' emotional tendencies and personal preferences.Therefore,current scholars are keen to conduct sentiment analysis on review text.However,user information and product information are important data in review websites.How to integrate user information and product information into the sentiment analysis model is a hot issue.The core research content of this article is to use deep learning methods to build a document-level sentiment analysis model that combines user information and product information for review text.main tasks as follows:(1)In order to capture text features more effectively,and at the same time integrate user information and product information into the sentiment analysis model,this paper proposes a model based on custom word embedding-bidirectional long short term memory network-attention mechanism.This model first integrates user information and product information into the word embedding module,and then combines the bidirectional long term short term memory network and the attention mechanism network layer to capture text features and semantic relationships.The experiment proves that model in this paper is effective.(2)In order to integrate user information and product information in a more diverse way,this paper proposes a model based on deep bidirectional long short term memory network-self attention mechanism-custom classifier.The model first uses a deep bidirectional long and short term memory network to identify contextual word meaning connections and obtain deep features of the text;then,it uses the self-attention mechanism network layer to capture important features in the text.The custom classifier module of the model integrates user information and product information,and uses the context-aware attention mechanism to prepare specific parameters for user information and product information to improve the analysis effect of the classifier module.The performance of this model on public data sets is better.The current common models have improved.(3)In order to better extract the textual information of word granularity,sentence granularity and chapter granularity contained in reviews,this paper proposes a hierarchical network model based on the attention interaction mechanism.The model uses a hierarchical network to extract semantic information of different granularities,and combines important information from users and products in the attention interaction mechanism to help extract text features.Finally,the loss value from the user's perspective and the loss value from the product's perspective are used as auxiliary analysis information,and the key text features of the user or product output by the hierarchical network are used for training and analysis.The performance of this model on the public data set is improved compared with the current common models.