Research On Short Text Sentiment Analysis Of Social Network Platform Based On Deep Learning

As the Internet continues to enter people’s daily lives,more and more people are willing to express their emotions on social platforms.There are a large number of short comment texts on social platforms,which play an important role in the study of public opinion and personal emotions in today’s society.Therefore,based on the characteristics of short Chinese texts,this paper proposes a new sentiment analysis algorithm model,and uses this model to analyze the emotional tendency of graduates in the comment texts about videos related to college students’ graduation season in Douyin APP.The main research content and innovations of this paper are as follows:(1)This paper proposes a MAOL model that combines an ordered neural longshort-term memory(ON-LSTM)network and a multi-head attention mechanism for short text sentiment classification,which solves the previous research on text sentiment analysis at home and abroad.Most of the network models used do not consider the information dependence of the long distance between words and only consider the local information between consecutive words,which may cause a part of the semantic information to be lost during the propagation process.The obtained short text emotional semantic information is more abundant,and the semantic features are more precise.The design experiment found that the classification accuracy of MAOL for short text reached 80.04%,and the performance was improved by more than 1.77 percentage points compared with other compared models,which proved that the MAOL model is efficient in the application of short text sentiment analysis.(2)This paper makes improvements on the MAOL model.On this basis,it combines the multi-channel RCNN model with the long-short-term memory(BiLSTM)network,and proposes a short text sentiment analysis model RCNN-MAOLBiLS,which uses tag semantics to expand text to enrich tag information,uses the MAOL model to integrate bidirectional memory neural network(Bi-LSTM)to obtain tag semantic information and uses a multi-channel RCNN model to obtain richer local features,which solves most of the problems.Text sentiment analysis models use sentiment categories as supervisory information in the classification stage,while ignoring the semantic information of labels to a certain extent.At the same time,model comparison experiments and ablation experiments were designed to find that the classification accuracy of this model continued to increase by 1.45% on the basis of the excellent performance of the MAOL model,and the RCNN-MAOL-BiLS model was successfully verified to further improve the performance of MAOL.(3)Design the system and apply the RCNN-MAOL-BiLS model proposed in this paper to the system,perform data preprocessing on the acquired data set,Bert pre-training and RCNN-MAOL-BiLS model training for emotion classification,and use the graph This paper presents the analysis results and puts forward relevant suggestions on the emotional orientation of college graduates reflected in the data set,so as to help graduates resist graduation pressure.