Chinese Sentiment Classification Algorithm Of Neural Network Based On Chaotic Lion Swarm Optimization

The rapid rise of Internet platform provides a place for people to express their views.The comment text contains rich subjective emotional tendencies.The research on analyzing the sentiment polarity of the text has attracted extensive attention.At present,with its powerful feature extraction ability,deep learning method has become the mainstream method of text sentiment analysis.However,the feature extraction ability of some basic models is single,resulting in incomplete feature representation.At the same time,the complex model structure is followed by a cumbersome parameter adjustment process.To address the problems existing in the deep learning method,taking the text emotion classification task as the starting point,this thesis constructs an sentiment analysis model based on chaotic lion group optimization and a multi-scale semantic collaborative network for text sentiment analysis.The main contributions of this thesis are as follows:(1)The traditional static word vector model can not be combined with the specific context of words for dynamic coding.Each word is represented by a unique vector,which can not represent polysemy.Dynamic word vector models ALBERT(A Lite BERT)and Ro BERTa(Robustly BERT Approach)are used to dynamically obtain text feature vectors,improve the accuracy of semantic representation of word vectors,and lay a foundation for subsequent semantic feature extraction.(2)Recurrent neural network is good at modeling sentences into sequences and capturing contextual semantic relevance,but it lacks the learning of the dependency between words and the structural features of sentences,resulting in the weak ability of feature capture.And because of the recurrent mechanism,the efficiency of parallel computing is low.To solve the above problems,this thesis proposes a simple recurrent unit with built-in fast attention(Bi FASRU),which captures the text syntax structure information through built-in fast attention and learns a higher-level abstract representation.The simple recurrent unit has fast parallel operation speed and reduces the training time and cost.(3)In order to further improve the comprehensiveness of the model feature representation,a multi-scale semantic collaborative network is constructed.The multi-scale convolution module is used to capture the local semantic features at different scales and input into the Bi FASRU module to model the text sequence,so as to enhance the semantic analysis ability of the model.(4)To address the problem that it is difficult to determine the super parameters such as the number of hidden layer units and batch size of Bi FASRU module,the global optimization ability of chaotic lion swarm algorithm is used to optimize the parameters to avoid complex and cumbersome parameter adjustment.