Research On Text Classification Of Government-people Interactive Messages Based On Deep Learning

In recent years,the advancement of science and technology has promoted the development of "Internet + government services",more and more government-people interactive platforms have entered the public’s field of vision.These platforms are important bridges for communication between the government and the public,through which the public can express their voices,and the government can understand people’s opinions and gather the wisdom of the people.However,with the advent of the era of big data,the number of messages on the government-people interactive platforms has also increased rapidly.How to quickly classify these data,so as to realize the effective interaction of information between the government and the public has become an urgent problem to be solved.In this thesis,taking some real messages of a government-people interactive platform in a province as research object,the BERT-BGMHA（BERT-Bi GRU-Multi-Head Attention）model was constructed to classify the text of the government-people interactive messages.Firstly,the text of the message was preprocessed,including de-weight,noise reduction and other operations.Secondly,the BERT medel was used to vectorize the preprocessed message text,and it’s effectiveness was proved by comparing with Word2 vec and Glove methods.Thirdly,constructed the BERT-BGMHA model to classify the message text,inputed the obtained message text vector into the Bi GRU layer for feature extraction,and then used the multi-head attention layer to capture important features in multiple vector spaces.Finally,the classification results were compared with those of other models,and each layer of the model was analyzed through ablation experiment.The experimental results show that the BERT-BGMHA model constructed in this thesis has achieved better classification results in this task,with the precision of 93.87%,the recall of 93.83%,and the F1 score of 93.83%,each layer of the model plays a positive role in improving the performance of the model.The text classification model BERT-BGMHA constructed in this thesis not only provides a scientific classification method for informal texts in the field of Chinese government affairs,but also plays a certain role in promoting the efficiency and governance of the government,enhancing the happiness of the public and maintaining social harmony and stability.