摘要：

Abstract:

【Purpose/significance】This paper proposes a model based on word vector and bi-directional GRU recurrent neural network

to improve the accuracy of sentiment classification of networked short text, which helps to pay attention to the emotional state of the

public on the network, maintain social stability, purify the network environment, and enhance people's happiness.【Method/process】

FastText algorithm was used to generate word vector and word vector, and the training effect of them in bidirectional GRU recurrent

neural network was compared to predict the sentiment classification of microblog comments.【Result/conclusion】The results show that

the use of word vector training can reduce the risk of model over fitting. The score of accuracy, precision, recall rate and F1 score of

the model proposed in this paper is above 0.92, which has excellent fitting ability and generalization ability.【Innovation/limitation】In

this paper, a unique word embedding layer and a cyclic neural network layer are configured for the model according to the theory. The

model performs well in the second category sentiment analysis task of Chinese short text, but its performance in the long text or three

category sentiment analysis task is unknown.