Research On Sentiment Analysis For Social Text Based On Deep Learning Model

With the outbreak of the Internet,major social networking sites,microblogging,and e-commerce platforms generate large amounts of user data every day,and the era of big data has come.Internet users can publish their own comments in the online world,their views on certain things,and their own attitudes.The Internet generates huge amounts of data each day to provide the basis for the algorithm.And,recently,the rapid development of hardware has enabled complex algorithms to be implemented.On the other hand,artificial intelligence has become increasingly hot in the past two years and has risen to the height of national strategy.Some interesting applications have been implemented in the fields of image,speech,natural language processing,and so on.Among the NLP tasks,how to correctly identify human emotion information is whether the machine can understand human intentions,whether it can have a better user experience,and whether it can provide the key to providing intimate services for humans.Then the user's sentiment analysis is particularly important..We refer to information published by users on the Internet platform or information that interacts with them as social text information.Usually,these social texts published by users contain personal emotions,preferences,tendencies,etc.Therefore,the sentiment analysis of social texts is a hot research topic.In several common models of deep learning,Recurrent Neural Network(RNN)is capable of handling large amounts of data.It is a widely used method for solving serial tasks,and because of its high accuracy,it is also a commonly used method in text sentiment analysis.After studying the related papers of the recurrent neural network RNN sentiment analysis,we found that the Long-Short Term Memory(LSTM)can capture the context-dependent semantic information,but it ignores the contextual emotional information;the sentence is used as a network.The input of the model can sometimes not be able to make a more accurate judgment of the sentimental tendency of the subject or the evaluation target of the sentence;when the text comment contains multiple aspects of the entity,it cannot accurately perform sentiment analysis for each entity.To solve the problems in the above three text sentiment analysis,this paper proposes 3 solutions in sequence:1.The standard LSTM can accurately identify the semantic information of texts,but the semantic information is not equal to sentiment information.For the sentiment information in the text,the Senti-LSTM model is proposed to make full use of the text context,semantic emotion and other information.2.Convolutional neural networks have more advantages in capturing local information.Recurrent neural networks are often used to solve sequential tasks.Combining convolutional long-term and short-term memory networks(CLSTM)is an effective method.This paper convolves long-term and short-term memory networks.The At-CLSTM(Attention Convolutional Long-Short Term Memory)model is integrated with the attention mechanism to capture the important information in the text.3.Text reviews often contain many different aspects,and for each different aspects of the sentiment tendencies may be different,this paper proposes an emotional analysis model that memorizes the long-term and short-term memory network aspect-level reviews,the model first identifies the text evaluation Different aspects are extracted,and based on the syntactic dependency relationship,the entity of the aspect and the emotion are modified to form a structured information,then this structured information is vectorized and finally processed using a long-term and short-term memory network..This article uses a number of real data sets containing Chinese and English for experiments to prove the effectiveness of the above three solutions.