Research on deep learning model of Chinese semantic classification

The short-term and long-term memory network divides English sentences into single words and numbers them according to the order in which the words appear. Due to the different order of words, the sequence number of each word is different. In semantic matching, the values of words that appear in training are random. This has little effect on English semantic matching. When used in Chinese, Chinese characters have a Chinese character library, and the location of each Chinese character in the Chinese character library is fixed. Therefore, we use the location code to compute the values of Chinese characters. Chinese characters and values are fixed values, and the semantics of Chinese characters and values are also one-to-one correspondence. When analyzing the semantics of Chinese characters, most researchers regard Chinese short sentences as the basis of semantic judgment, but the division of short sentences is difficult. Improper segmentation will damage the original semantics. Using LSTM algorithm for reference, we build a four layer CNN neural network and develop our own deep learning algorithm. Firstly, the sentence is cut according to the word, and then the location code is computed as the value of Chinese characters and sent to the neural network for training. Experiments show that after these two improvements, both LSTM and CNN algorithms can adapt to Chinese semantic matching and achieve good matching results.