Sentiment Analysis Based On The Framework Of Stacking And Weakly Supervised Deep Learning

With the rapid development of the Internet,Internet social media has become the main platform for people to obtain information,present opinions and express emotions.These emotions and opinions often contain a lot of valuable information,which has a subtle influence on various aspects such as social political and economic development.So how to study and use these emotional information has a great of significance.Current mainstream sentiment analysis methods include emotional dictionary combined with syntactic rules and the use of machine learning or deep learning methods to construct sentiment analysis models.The dictionary method relies on the construction of sentiment dictionary,and it is necessary to absorb new words,irregular vocabulary and morphological words in time.In the meantime,there are problems such as high dictionary update requirements and insufficient precision.The sentiment analysis using machine learning and deep learning depends on the construction of the model and the selection of features.It is necessary to provide an effective training library when training the model.Traditional supervised learning algorithms need to tag data,but with the development of the times,the emergence of a large number of untagged data makes the tag cost higher and higher,which is almost unacceptable when dealing with massive data.In response to this problem,this paper proposes to use the idea of weakly supervised learning method to predict untagged data by using pre-trained models in the case of a small number of tags.Then the prediction result is re-entered into the model as training data for training,and constantly improve the sentiment classification performance of the model in order to solve the high tag cost and the insufficient training data volume problems.In view of the fact that the current mainstream machine learning and deep learning sentiment analysis algorithms have their main areas of expertise,but also have their own shortcomings,such as the Na?veBayes classification model algorithm,which is simple and the time space overhead is small,but the relationship of sample attributes cannot be learned.The support vector machine is mainly good at solving the problem of binary classification pattern recognition.Long short-term memory networks have good performance when analyzing sequence data,but their performance is not ideal in extracting emotional features in parallel.Convolutional neural networks have strong structural feature capture capabilities,but can not find correlations between sequences.In response to this problem,this paper proposes to combine various machine learning and deep learning algorithms by using the Stacking framework in ensemble learning.Combine the weakly supervised learning method to achieve the goal of improving the performance of sentiment analysis as a whole by making full use of the advantages of each classification model and making up for its shortcomings in the case of a small number of tagged training libraries.The main contributions of this paper are as follows:(1)In view of the lack of emotional training libraries and the problem of excessive cost tagging of massive data,it is proposed to use the weakly supervised learning method to continuously improve the emotional classification performance of the model in the case of a small number of markers.(2)Aiming at the advantages and disadvantages of the mainstream machine learning and deep learning sentiment analysis models,the problem of the performance bottleneck of a single classifier model is used,it is proposed to use the Stacking framework in ensemble learning to combine different models and combine the weakly supervised learning method to further improve the performance of sentiment analysis models in the absence of training libraries.(3)This paper will put the relevant code and data of the weakly supervised deep learning sentiment analysis model based on the Stacking framework as an open source project on the GitHub platform,for researchers in related fields to further explore the sentiment analysis technology.