Text Sentiment Polarity Classification Method Based on Word Embedding

Most of the machine learning algorithms for text sentiment analysis use the word embedding obtained by word2vec training as their inputs. However, the word embedding of word2vec training contains only semantic information. An algorithm for text sentiment analysis is proposed solve the problem of text containing semantics, syntax, sentiment and other information. It begins with the learning of original text-multi word embedding in the semantic, syntactic, and sentiment information, followed by proceeding the word embedding fusion. The improved convolution neural network is applied for sentiment analysis. Thus, it solves the problem that the word embedding contains monotonous text information. K-means text clustering is applied by dividing similar text into the same cluster, thus improving the classification accuracy. The application of the Principal Component Analysis (PCA) dimensionality not only extracts the principal component information, but also solves the problem of redundancy embedding and improves the computational performance of classification model. The experiment results show that the presented method has a significant improvement in the accuracy, recall rate and F value of the sentiment polarity analysis of the critical text in comparison with other fusion algorithms.