Research On The Application Of Sentiment Analysis Of Weibo Comments Based On Deep Learning

In recent years,social applications led by Weibo have achieved rapid development driven by the Internet.The number of Weibo users has rapidly increased from over one million at the beginning to nearly one billion now,and the vast amount of comment data generated by Weibo cannot be underestimated.Through conducting sentiment analysis on these data,it is possible to gain more accurate understanding of people’s views on trending topics,commercial products,and public figures,which provides a real-time and scientific basis for decision-making on public opinion regulation,marketing,and public relations crisis.This thesis focuses on sentiment analysis of Weibo comments,and the main works are as follows:1)The pre-trained model BERT is used in this thesis for feature extraction to obtain contextually relevant dynamic word vectors.This approach is different from the traditional method of obtaining word vectors using Word2 Vec.Instead,this approach uses the SelfAttention mechanism within bidirectional deep Transformer model for training word vectors.The word vectors generated by BERT not only contain contextual information,but can also be adaptively adjusted in different language contexts,overcoming the limitation of representing a single meaning in traditional word vectors.2)To address the problem of inadequate feature extraction in pre-trained models,a bidirectional long and short-term memory network model based on BERT feature extraction,the BERT-BiLSTM model,is proposed.The pre-trained model BERT is trained on largescale datasets to obtain a general feature representation,but the general feature representation is not sufficient for the specific task requirements of this thesis.Therefore,in order to obtain more fine-grained feature,on the basis of the above word vector method,BERT’s output vectors are spliced to obtain a vector matrix,and then the vector matrix is used as the input of the bidirectional long and short-term memory network BiLSTM.Using the features of BiLSTM,the word vector features are further filtered to obtain a more finegrained feature representation.3)In order to further improve the classification accuracy of the model,a BERTBiLSTM-Attention model with Attention is proposed.BiLSTM outputs feature vectors for each hidden state,and uses Attention to allocate resources based on weights.The hidden state vectors with different levels of importance are weighted and summed to obtain the feature matrix for classification.4)Finally,the sentiment analysis model in this thesis is combined with practical applications to build a sentiment analysis system based on deep learning.The system is capable of automatically identifying and analyzing the emotional state in the text,thereby assisting users in making informed decisions.To build and visualise the system,frameworks such as Vue,Django,and Scrapyd are used,and the implementation be done in Python.Additionally,Echarts is used for visual representation.