| Under the background that traditional industries use the "Internet +" innovation model for industrial upgrading,a large enterprise in Zhejiang Province has used Internet technology to realize the construction of an "Internet + Marketing" information platform.Since the official launch of the marketing platform,users on the platform have the review data of marketing products has achieved explosive growth.How to mine users’ feedback on products from these text data and assist enterprises to upgrade and improve products is a key issue that companies are currently concerned about.Therefore,under the support of the comment data of millions of users of the marketing platform,this paper proposes a deep learning network algorithm that integrates prior knowledge and a text sentiment analysis method based on the improved BERT dual-channel neural network algorithm.Analysis,mining users’ opinions on marketing products,and finally realizing the application of the proposed algorithm model in combination with the Internet marketing platform.The main research contents of this paper are summarized as follows:1)Aiming at the problem that the construction cost of sentiment dictionary is high and the sentiment features contained in the vectorized word vectors are insufficient,a deep learning text sentiment analysis algorithm integrating prior knowledge is proposed.Firstly,the method adopts the vector cosine similarity calculation and SO-PMI algorithm to realize the automatic construction of the domain sentiment dictionary,and then uses the domain sentiment dictionary as a prior knowledge base to enhance the sentiment of the text sequence.Secondly,the Word2 Vec word embedding model is used to obtain the sentiment-enhanced text word vector sequence,and the bidirectional gated recurrent unit network and the self-attention network are used on the feature extraction layer to capture the text sentiment features.Experiments show that the proposed algorithm model can capture text emotional features more efficiently than other benchmark models.2)Aiming at the problems that the static word embedding model cannot effectively represent polysemy words in the current sentiment analysis task,and the single-layer deep learning algorithm cannot fully capture the text sentiment features,a text sentiment analysis algorithm based on improved BERT is proposed.This method uses the full-word masked BERT model as the word embedding layer,and combines the convolutional neural network and the bidirectional gated recurrent unit network to fo h rm a two-channel feature extraction layer.The output is based on the importance of the emotional features,and the local and global features are assigned corresponding weight scores through the self-attention network to highlight the emotional feature information captured by the feature extraction layer.The comparative experiments show that the dynamic word vector obtained by using the BERT model combined with the dual-channel neural network model structure can more effectively judge the sentimental tendency of the text.3)The application of user comment sentiment analysis algorithm in Internet marketing platform is realized.Firstly,the background and system architecture of the Internet marketing platform are summarized.Secondly,the platform text sentiment analysis module is designed and implemented.The bottom layer of this module relies on the data service capabilities provided by the big data infrastructure platform,and uses Python to implement the user comment sentiment analysis algorithm.For model training and invocation,the microservice architecture is used to develop the sentiment analysis service interface.Each business microservice requests the sentiment analysis interface through remote calls.The analysis results are visualized and developed using the frontend framework Vue.js and middleware ECharts.The Internet marketing platform relies on the text sentiment analysis module to complete the real-time analysis of user comment data. |
