Deep Learning Based Text Sentiment Analysis

Statistical learning,which essentially takes data as the research object,uses computer,mathematics,probability theory and other interdisciplinary knowledge to build a model,and it finally extracts the features to analyze data.Text sentiment analysis,as a major research content in the field of statistical learning,takes the unstructured text data left by users in various media as the research object,and statistically analyzes text data to extract the valuable structural information,at present,text sentiment analysis has been widely used in public opinion monitoring,financial analysis,commodity recommendation and so on.In recent years,deep learning has achieved good results in processing various natural language processing tasks such as part-of-speech tagging,abstract extraction,and text classification.Deep learning makes the data in a distributed representation and constructs a complex network structure to learn the deep abstract features of data,which avoids the tedious feature construction work and enables the model to have good generalization capabilities.Therefore,this paper mainly studies the deep learning model and applies it to the sentiment analysis of web reviews.The specific research content is as follows:(1)Two kinds of traditional sentiment analysis methods based on lexicography and machine learning are introduced,and their algorithm flow,advantages,and disadvantages are analyzed in detail.(2)The basic principles of deep learning models such as convolutional neural networks and long short-term memory networks are described,and a sentiment analysis model combining convolutional neural network and long short-term memory network is proposed.This model uses the trained word vector to represent the text and adopts the parallel two-layer convolutional network to learn the local features of the text.At the same time,it stacks the long short-term memory networks to learn the hidden sequence features of the text.Experimental results show that the proposed model shows better performance compared with traditional machine learning methods and several deep learning models.(3)Based on(2),this paper also proposes a new sentiment analysis model.This model mainly uses multiple sub-training sets and sub-models to extract more detailed features of individual categories.The extracted features have a strong characterization contribution to the determination of the individual categories.The experimental results show that the model shows good performance in three kinds of emotional discrimination tasks:positive,negative,and neutral.