An improved method of sentence classification which combines depth learning and mathematical analysis

The invention provides an improved method for sentence classification which combines depth learning and mathematical analysis. The method combines the advantages of deep learning and mathematical analysis in dealing with sentence problems, that is, long-term and short-term memory (LSTM) network can take into account the word order information and context information of words in sentences, the inverse word frequency weight (AWF) can highlight the statistical features of words in the corpus. By subtracting the projection of S0 from the original vector representation S0 on the first principal component V1, the improved sentence feature vector representation S1 is obtained, and the sentence classification result is obtained by using S1 as the input of softmax layer. Combining these advantagestogether and learning from each other is helpful for the reliability of sentence modeling to get a better representation of sentence semantic features, thus improving the accuracy of sentence classification, which can also be